diff --git a/backend/internal/api/admin_channels.go b/backend/internal/api/admin_channels.go
new file mode 100644
index 0000000..4ebb1b2
--- /dev/null
+++ b/backend/internal/api/admin_channels.go
@@ -0,0 +1,588 @@
+package api
+
+import (
+	"encoding/json"
+	"errors"
+	"fmt"
+	"io"
+	"net/http"
+	"strconv"
+	"strings"
+	"time"
+
+	"opencatd-open/internal/pkg/crypto"
+	"opencatd-open/internal/store"
+
+	"github.com/gin-gonic/gin"
+)
+
+// AdminChannels GET /api/admin/channels — 渠道列表（不返回加密 key，返回掩码）。
+func (h *Handler) AdminChannels(c *gin.Context) {
+	var chs []store.Channel
+	if err := h.db.Order("id ASC").Find(&chs).Error; err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to load channels"})
+		return
+	}
+	out := make([]gin.H, 0, len(chs))
+	for _, ch := range chs {
+		masked := ""
+		if key, err := crypto.Decrypt(ch.APIKeyEnc); err == nil && len(key) > 8 {
+			masked = maskAPIKey(key)
+		} else if err == nil {
+			masked = "****"
+		}
+		out = append(out, gin.H{
+			"id": ch.ID, "name": ch.Name, "provider": ch.Provider, "formats": ch.FormatsEffective(),
+			"base_url": ch.BaseURL, "base_urls": ch.BaseURLs,
+			"api_key_masked": masked, "weight": ch.Weight, "priority": ch.Priority,
+			"timeout_ms": ch.TimeoutMS, "max_concurrency": ch.MaxConcurrency,
+			"health_status": ch.HealthStatus, "enabled": ch.Enabled,
+			"created_at": ch.CreatedAt,
+		})
+	}
+	c.JSON(http.StatusOK, gin.H{"data": out})
+}
+
+type channelBody struct {
+	Name           string            `json:"name" binding:"required,min=1,max=64"`
+	Provider       string            `json:"provider"`
+	Formats        []string          `json:"formats"`
+	BaseURL        string            `json:"base_url"`
+	BaseURLs       map[string]string `json:"base_urls"`
+	APIKey         string            `json:"api_key"`
+	Weight         *int              `json:"weight"`
+	Priority       *int              `json:"priority"`
+	TimeoutMS      *int              `json:"timeout_ms"`
+	MaxConcurrency *int              `json:"max_concurrency"`
+	Enabled        *bool             `json:"enabled"`
+}
+
+// normalizeBaseURLs 校验并清理分协议 base_url。
+func normalizeBaseURLs(m map[string]string) map[string]string {
+	if len(m) == 0 {
+		return nil
+	}
+	out := map[string]string{}
+	for k, v := range m {
+		if validFormats[k] && strings.TrimSpace(v) != "" {
+			out[k] = strings.TrimRight(strings.TrimSpace(v), "/")
+		}
+	}
+	if len(out) == 0 {
+		return nil
+	}
+	return out
+}
+
+// resolveBaseURL 渠道 base_url：留空按供应商默认；网关按内容智能识别前缀/完整端点。
+func resolveBaseURL(provider, raw string) (string, error) {
+	base := strings.TrimRight(raw, "/")
+	if base == "" {
+		switch provider {
+		case store.ChannelProviderOpenAI:
+			base = "https://api.openai.com"
+		case store.ChannelProviderAnthropic:
+			base = "https://api.anthropic.com"
+		}
+	}
+	if base == "" {
+		return "", errors.New("base_url required for compatible channels")
+	}
+	return base, nil
+}
+
+func validateProvider(p string) bool {
+	return p == store.ChannelProviderOpenAI || p == store.ChannelProviderAnthropic || p == store.ChannelProviderCompatible
+}
+
+var validFormats = map[string]bool{
+	store.FormatChat: true, store.FormatResponses: true, store.FormatMessages: true,
+}
+
+// deriveProvider 按格式推断供应商（仅作内部字段/兼容用途，不参与路由）。
+func deriveProvider(formats []string) string {
+	if len(formats) == 0 {
+		return store.ChannelProviderCompatible
+	}
+	messagesOnly, hasResponses := true, false
+	for _, f := range formats {
+		if f != store.FormatMessages {
+			messagesOnly = false
+		}
+		if f == store.FormatResponses {
+			hasResponses = true
+		}
+	}
+	if messagesOnly {
+		return store.ChannelProviderAnthropic
+	}
+	if hasResponses {
+		return store.ChannelProviderOpenAI
+	}
+	return store.ChannelProviderCompatible
+}
+
+// resolveFormats 渠道协议格式：显式给出则校验去重；空则按 provider 推断默认。
+func resolveFormats(provider string, formats []string) ([]string, error) {
+	if len(formats) == 0 {
+		switch provider {
+		case store.ChannelProviderAnthropic:
+			return []string{store.FormatMessages}, nil
+		case store.ChannelProviderOpenAI:
+			return []string{store.FormatChat, store.FormatResponses}, nil
+		default:
+			return []string{store.FormatChat}, nil
+		}
+	}
+	seen := map[string]bool{}
+	out := make([]string, 0, len(formats))
+	for _, f := range formats {
+		if !validFormats[f] {
+			return nil, fmt.Errorf("unsupported format %q", f)
+		}
+		if !seen[f] {
+			seen[f] = true
+			out = append(out, f)
+		}
+	}
+	return out, nil
+}
+
+// AdminCreateChannel POST /api/admin/channels
+func (h *Handler) AdminCreateChannel(c *gin.Context) {
+	var req channelBody
+	if err := c.ShouldBindJSON(&req); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid input: " + err.Error()})
+		return
+	}
+	if req.Provider == "" {
+		req.Provider = deriveProvider(req.Formats)
+	}
+	if !validateProvider(req.Provider) {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "provider must be openai, anthropic or compatible"})
+		return
+	}
+	if req.APIKey == "" {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "api_key required"})
+		return
+	}
+	formats, err := resolveFormats(req.Provider, req.Formats)
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
+		return
+	}
+	baseURL, err := resolveBaseURL(req.Provider, req.BaseURL)
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
+		return
+	}
+	enc, err := crypto.Encrypt(req.APIKey)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to encrypt api key"})
+		return
+	}
+	ch := store.Channel{
+		Name: req.Name, Provider: req.Provider, Formats: formats, BaseURL: baseURL,
+		BaseURLs: normalizeBaseURLs(req.BaseURLs),
+		APIKeyEnc: enc, Weight: intOr(req.Weight, 1), Priority: intOr(req.Priority, 0),
+		TimeoutMS: intOr(req.TimeoutMS, 120000), MaxConcurrency: intOr(req.MaxConcurrency, 16),
+		HealthStatus: store.ChannelHealthHealthy, Enabled: boolOr(req.Enabled, true),
+	}
+	if err := h.db.Create(&ch).Error; err != nil {
+		c.JSON(http.StatusConflict, gin.H{"error": "failed to create channel (name may already exist)"})
+		return
+	}
+	c.JSON(http.StatusCreated, gin.H{"id": ch.ID, "name": ch.Name})
+}
+
+// AdminUpdateChannel PUT /api/admin/channels/:id
+func (h *Handler) AdminUpdateChannel(c *gin.Context) {
+	id, err := strconv.ParseUint(c.Param("id"), 10, 64)
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid channel id"})
+		return
+	}
+	var body struct {
+		Name           *string            `json:"name"`
+		Provider       *string            `json:"provider"`
+		Formats        *[]string          `json:"formats"`
+		BaseURL        *string            `json:"base_url"`
+		BaseURLs       *map[string]string `json:"base_urls"`
+		APIKey         *string            `json:"api_key"`
+		Weight         *int               `json:"weight"`
+		Priority       *int               `json:"priority"`
+		TimeoutMS      *int               `json:"timeout_ms"`
+		MaxConcurrency *int               `json:"max_concurrency"`
+		HealthStatus   *string            `json:"health_status"`
+		Enabled        *bool              `json:"enabled"`
+	}
+	if err := c.ShouldBindJSON(&body); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid input"})
+		return
+	}
+	var ch store.Channel
+	if err := h.db.First(&ch, id).Error; err != nil {
+		c.JSON(http.StatusNotFound, gin.H{"error": "channel not found"})
+		return
+	}
+	updates := map[string]any{}
+	if body.Name != nil {
+		updates["name"] = *body.Name
+	}
+	if body.Provider != nil {
+		if !validateProvider(*body.Provider) {
+			c.JSON(http.StatusBadRequest, gin.H{"error": "provider must be openai, anthropic or compatible"})
+			return
+		}
+		updates["provider"] = *body.Provider
+	}
+	if body.BaseURL != nil {
+		prov := ch.Provider
+		if body.Provider != nil {
+			prov = *body.Provider
+		}
+		b, berr := resolveBaseURL(prov, *body.BaseURL)
+		if berr != nil {
+			c.JSON(http.StatusBadRequest, gin.H{"error": berr.Error()})
+			return
+		}
+		updates["base_url"] = b
+	}
+	if body.BaseURLs != nil {
+		raw, _ := json.Marshal(normalizeBaseURLs(*body.BaseURLs))
+		updates["base_urls"] = string(raw)
+	}
+	if body.APIKey != nil && *body.APIKey != "" {
+		enc, err := crypto.Encrypt(*body.APIKey)
+		if err != nil {
+			c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to encrypt api key"})
+			return
+		}
+		updates["api_key_enc"] = enc
+	}
+	if body.Weight != nil {
+		updates["weight"] = *body.Weight
+	}
+	if body.Priority != nil {
+		updates["priority"] = *body.Priority
+	}
+	if body.TimeoutMS != nil {
+		updates["timeout_ms"] = *body.TimeoutMS
+	}
+	if body.MaxConcurrency != nil {
+		updates["max_concurrency"] = *body.MaxConcurrency
+	}
+	if body.HealthStatus != nil {
+		updates["health_status"] = *body.HealthStatus
+	}
+	if body.Enabled != nil {
+		updates["enabled"] = *body.Enabled
+	}
+	if body.Formats != nil {
+		prov := ch.Provider
+		if body.Provider != nil {
+			prov = *body.Provider
+		}
+		formats, ferr := resolveFormats(prov, *body.Formats)
+		if ferr != nil {
+			c.JSON(http.StatusBadRequest, gin.H{"error": ferr.Error()})
+			return
+		}
+		raw, _ := json.Marshal(formats)
+		updates["formats"] = string(raw)
+	}
+	if len(updates) > 0 {
+		if err := h.db.Model(&ch).Updates(updates).Error; err != nil {
+			c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to update channel"})
+			return
+		}
+	}
+	c.JSON(http.StatusOK, gin.H{"ok": true})
+}
+
+// AdminDeleteChannel DELETE /api/admin/channels/:id
+func (h *Handler) AdminDeleteChannel(c *gin.Context) {
+	id, err := strconv.ParseUint(c.Param("id"), 10, 64)
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid channel id"})
+		return
+	}
+	res := h.db.Delete(&store.Channel{}, id)
+	if res.Error != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to delete channel"})
+		return
+	}
+	if res.RowsAffected == 0 {
+		c.JSON(http.StatusNotFound, gin.H{"error": "channel not found"})
+		return
+	}
+	h.db.Where("channel_id = ?", id).Delete(&store.ChannelModelBinding{})
+	c.JSON(http.StatusOK, gin.H{"ok": true})
+}
+
+// AdminTestChannel POST /api/admin/channels/:id/test — 请求渠道 /v1/models 测连通性。
+func (h *Handler) AdminTestChannel(c *gin.Context) {
+	id, err := strconv.ParseUint(c.Param("id"), 10, 64)
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid channel id"})
+		return
+	}
+	var ch store.Channel
+	if err := h.db.First(&ch, id).Error; err != nil {
+		c.JSON(http.StatusNotFound, gin.H{"error": "channel not found"})
+		return
+	}
+	key, err := crypto.Decrypt(ch.APIKeyEnc)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to decrypt channel key"})
+		return
+	}
+	url := ch.UpstreamURL("", "/models")
+	client := &http.Client{Timeout: 10 * time.Second}
+	req, _ := http.NewRequest(http.MethodGet, url, nil)
+	req.Header.Set("Authorization", "Bearer "+key)
+	req.Header.Set("Accept", "application/json")
+	start := time.Now()
+	resp, err := client.Do(req)
+	status := store.ChannelHealthHealthy
+	msg := "ok"
+	latency := 0
+	if err != nil {
+		status = store.ChannelHealthCooldown
+		msg = err.Error()
+	} else {
+		latency = int(time.Since(start).Milliseconds())
+		if resp.StatusCode < 200 || resp.StatusCode >= 300 {
+			status = store.ChannelHealthCooldown
+			b, _ := io.ReadAll(io.LimitReader(resp.Body, 1024))
+			msg = fmt.Sprintf("http %d: %s", resp.StatusCode, strings.TrimSpace(string(b)))
+		}
+		resp.Body.Close()
+	}
+	h.db.Model(&store.Channel{}).Where("id = ?", ch.ID).Update("health_status", status)
+	if status != store.ChannelHealthHealthy {
+		c.JSON(http.StatusBadGateway, gin.H{"error": msg})
+		return
+	}
+	c.JSON(http.StatusOK, gin.H{"ok": true, "latency_ms": latency, "message": msg})
+}
+
+// AdminChannelRemoteModels GET /api/admin/channels/:id/models/remote — 拉取远端模型列表。
+// 返回本渠道尚未允许的模型（新增候选），排除已绑定的模型。
+func (h *Handler) AdminChannelRemoteModels(c *gin.Context) {
+	id, err := strconv.ParseUint(c.Param("id"), 10, 64)
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid channel id"})
+		return
+	}
+	var ch store.Channel
+	if err := h.db.First(&ch, id).Error; err != nil {
+		c.JSON(http.StatusNotFound, gin.H{"error": "channel not found"})
+		return
+	}
+	key, err := crypto.Decrypt(ch.APIKeyEnc)
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to decrypt channel key"})
+		return
+	}
+	url := ch.UpstreamURL("", "/models")
+	client := &http.Client{Timeout: 10 * time.Second}
+	req, _ := http.NewRequest(http.MethodGet, url, nil)
+	req.Header.Set("Authorization", "Bearer "+key)
+	req.Header.Set("Accept", "application/json")
+	resp, err := client.Do(req)
+	if err != nil {
+		c.JSON(http.StatusBadGateway, gin.H{"error": err.Error()})
+		return
+	}
+	defer resp.Body.Close()
+	body, _ := io.ReadAll(io.LimitReader(resp.Body, 10*1024*1024))
+	if resp.StatusCode < 200 || resp.StatusCode >= 300 {
+		c.JSON(http.StatusBadGateway, gin.H{"error": fmt.Sprintf("http %d: %s", resp.StatusCode, string(body))})
+		return
+	}
+	// 解析 OpenAI 格式的模型列表
+	var result struct {
+		Data []struct {
+			ID string `json:"id"`
+		} `json:"data"`
+	}
+	if err := json.Unmarshal(body, &result); err != nil {
+		c.JSON(http.StatusBadGateway, gin.H{"error": "failed to parse response: " + err.Error()})
+		return
+	}
+	// 本渠道已允许的上游模型名：不作为新增候选
+	var boundNames []string
+	h.db.Model(&store.ChannelModelBinding{}).Where("channel_id = ?", id).Pluck("upstream_model", &boundNames)
+	boundSet := make(map[string]bool, len(boundNames))
+	for _, n := range boundNames {
+		boundSet[strings.TrimSpace(n)] = true
+	}
+	models := make([]string, 0, len(result.Data))
+	for _, m := range result.Data {
+		name := strings.TrimSpace(m.ID)
+		if name != "" && !boundSet[name] {
+			models = append(models, name)
+		}
+	}
+	c.JSON(http.StatusOK, gin.H{"data": models})
+}
+
+// AdminChannelModels GET /api/admin/channels/:id/models — 渠道绑定列表。
+func (h *Handler) AdminChannelModels(c *gin.Context) {
+	channelID, err := strconv.ParseUint(c.Param("id"), 10, 64)
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid channel id"})
+		return
+	}
+	var bindings []store.ChannelModelBinding
+	if err := h.db.Preload("Model").Where("channel_id = ?", channelID).Find(&bindings).Error; err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to load bindings"})
+		return
+	}
+	out := make([]gin.H, 0, len(bindings))
+	for _, b := range bindings {
+		out = append(out, gin.H{
+			"id": b.ID, "model_id": b.ModelID, "model_name": b.Model.Name,
+			"upstream_model": b.UpstreamModel, "weight": b.Weight,
+		})
+	}
+	c.JSON(http.StatusOK, gin.H{"data": out})
+}
+
+// AdminChannelAddModel POST /api/admin/channels/:id/models — 手工添加模型绑定。
+// 无需渠道具备 /v1/models 接口：直接填上游模型名，可选自定义名称作为客户端调用名。
+func (h *Handler) AdminChannelAddModel(c *gin.Context) {
+	channelID, err := strconv.ParseUint(c.Param("id"), 10, 64)
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid channel id"})
+		return
+	}
+	var req struct {
+		UpstreamModel string `json:"upstream_model" binding:"required"` // 渠道侧真实模型名
+		CustomName    string `json:"custom_name"`                      // 客户端调用名，空=用上游名
+		Weight        *int   `json:"weight"`
+	}
+	if err := c.ShouldBindJSON(&req); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid input: upstream_model required"})
+		return
+	}
+	globalName := req.CustomName
+	if globalName == "" {
+		globalName = req.UpstreamModel
+	}
+	var ch store.Channel
+	if err := h.db.First(&ch, channelID).Error; err != nil {
+		c.JSON(http.StatusNotFound, gin.H{"error": "channel not found"})
+		return
+	}
+	// 查找或创建全局模型
+	var m store.Model
+	if err := h.db.Where("name = ?", globalName).First(&m).Error; err != nil {
+		m = store.Model{Name: globalName, Enabled: true}
+		if err := h.db.Create(&m).Error; err != nil {
+			c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to create model"})
+			return
+		}
+	}
+	// 查找已存在的绑定，如果存在则更新
+	var existing store.ChannelModelBinding
+	if err := h.db.Where("channel_id = ? AND model_id = ?", channelID, m.ID).First(&existing).Error; err == nil {
+		// 已存在，更新
+		existing.UpstreamModel = req.UpstreamModel
+		existing.Weight = intOr(req.Weight, 1)
+		if err := h.db.Save(&existing).Error; err != nil {
+			c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to update binding"})
+			return
+		}
+		c.JSON(http.StatusOK, gin.H{"id": existing.ID, "model_id": m.ID, "model_name": m.Name, "upstream_model": existing.UpstreamModel, "weight": existing.Weight})
+		return
+	}
+	// 不存在，创建新的
+	b := store.ChannelModelBinding{
+		ChannelID: channelID, ModelID: m.ID,
+		UpstreamModel: req.UpstreamModel, Weight: intOr(req.Weight, 1),
+	}
+	if err := h.db.Create(&b).Error; err != nil {
+		c.JSON(http.StatusConflict, gin.H{"error": "binding may already exist"})
+		return
+	}
+	c.JSON(http.StatusCreated, gin.H{"id": b.ID, "model_id": m.ID, "model_name": m.Name, "upstream_model": req.UpstreamModel, "weight": b.Weight})
+}
+
+// AdminChannelUpdateModel PATCH /api/admin/channels/:id/models/:bid — 改映射名/权重。
+func (h *Handler) AdminChannelUpdateModel(c *gin.Context) {
+	bid, err := strconv.ParseUint(c.Param("bid"), 10, 64)
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid binding id"})
+		return
+	}
+	var req struct {
+		UpstreamModel *string `json:"upstream_model"`
+		Weight        *int    `json:"weight"`
+	}
+	if err := c.ShouldBindJSON(&req); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid input"})
+		return
+	}
+	var b store.ChannelModelBinding
+	if err := h.db.First(&b, bid).Error; err != nil {
+		c.JSON(http.StatusNotFound, gin.H{"error": "binding not found"})
+		return
+	}
+	updates := map[string]any{}
+	if req.UpstreamModel != nil {
+		updates["upstream_model"] = *req.UpstreamModel
+	}
+	if req.Weight != nil {
+		updates["weight"] = *req.Weight
+	}
+	if len(updates) > 0 {
+		if err := h.db.Model(&b).Updates(updates).Error; err != nil {
+			c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to update binding"})
+			return
+		}
+	}
+	c.JSON(http.StatusOK, gin.H{"ok": true})
+}
+
+// AdminChannelDeleteModel DELETE /api/admin/channels/:id/models/:bid — 解除绑定。
+func (h *Handler) AdminChannelDeleteModel(c *gin.Context) {
+	bid, err := strconv.ParseUint(c.Param("bid"), 10, 64)
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid binding id"})
+		return
+	}
+	res := h.db.Delete(&store.ChannelModelBinding{}, bid)
+	if res.Error != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to delete binding"})
+		return
+	}
+	if res.RowsAffected == 0 {
+		c.JSON(http.StatusNotFound, gin.H{"error": "binding not found"})
+		return
+	}
+	c.JSON(http.StatusOK, gin.H{"ok": true})
+}
+
+// maskAPIKey 掩码渠道密钥：保留前 7 位与后 4 位，中间固定 ****** 遮蔽。
+func maskAPIKey(key string) string {
+	if len(key) <= 11 {
+		return strings.Repeat("*", len(key)-4) + key[len(key)-4:]
+	}
+	return key[:7] + "******" + key[len(key)-4:]
+}
+
+func intOr(p *int, def int) int {
+	if p == nil {
+		return def
+	}
+	return *p
+}
+
+func boolOr(p *bool, def bool) bool {
+	if p == nil {
+		return def
+	}
+	return *p
+}
diff --git a/backend/internal/api/admin_config.go b/backend/internal/api/admin_config.go
new file mode 100644
index 0000000..95ab562
--- /dev/null
+++ b/backend/internal/api/admin_config.go
@@ -0,0 +1,109 @@
+package api
+
+import (
+	"net/http"
+
+	"opencatd-open/internal/store"
+
+	"github.com/gin-gonic/gin"
+)
+
+// AdminGetConfig GET /api/admin/config — 获取系统配置。
+func (h *Handler) AdminGetConfig(c *gin.Context) {
+	configs := map[string]string{}
+	var rows []store.SystemConfig
+	h.db.Find(&rows)
+	for _, r := range rows {
+		configs[r.Key] = r.Value
+	}
+	c.JSON(http.StatusOK, gin.H{"data": configs})
+}
+
+// AdminUpdateConfig PUT /api/admin/config — 更新系统配置。
+func (h *Handler) AdminUpdateConfig(c *gin.Context) {
+	var req map[string]string
+	if err := c.ShouldBindJSON(&req); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid input"})
+		return
+	}
+	for key, value := range req {
+		var sc store.SystemConfig
+		result := h.db.Where("key = ?", key).First(&sc)
+		if result.Error == nil {
+			sc.Value = value
+			h.db.Save(&sc)
+		} else {
+			sc = store.SystemConfig{Key: key, Value: value}
+			h.db.Create(&sc)
+		}
+	}
+	c.JSON(http.StatusOK, gin.H{"ok": true})
+}
+
+// AdminGetRegistration GET /api/admin/config/registration — 获取注册配置。
+func (h *Handler) AdminGetRegistration(c *gin.Context) {
+	var sc store.SystemConfig
+	enabled := "true"
+	if err := h.db.Where("key = ?", "registration_enabled").First(&sc).Error; err == nil {
+		enabled = sc.Value
+	}
+	c.JSON(http.StatusOK, gin.H{"data": gin.H{"enabled": enabled == "true"}})
+}
+
+// AdminUpdateRegistration PUT /api/admin/config/registration — 更新注册配置。
+func (h *Handler) AdminUpdateRegistration(c *gin.Context) {
+	var req struct {
+		Enabled bool `json:"enabled"`
+	}
+	if err := c.ShouldBindJSON(&req); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid input"})
+		return
+	}
+	value := "false"
+	if req.Enabled {
+		value = "true"
+	}
+	var sc store.SystemConfig
+	if err := h.db.Where("key = ?", "registration_enabled").First(&sc).Error; err == nil {
+		sc.Value = value
+		h.db.Save(&sc)
+	} else {
+		sc = store.SystemConfig{Key: "registration_enabled", Value: value}
+		h.db.Create(&sc)
+	}
+	c.JSON(http.StatusOK, gin.H{"ok": true})
+}
+
+// AdminGetPasswordLogin GET /api/admin/config/password-login — 获取密码登录配置。
+func (h *Handler) AdminGetPasswordLogin(c *gin.Context) {
+	var sc store.SystemConfig
+	enabled := "true"
+	if err := h.db.Where("key = ?", "password_login_enabled").First(&sc).Error; err == nil {
+		enabled = sc.Value
+	}
+	c.JSON(http.StatusOK, gin.H{"data": gin.H{"enabled": enabled == "true"}})
+}
+
+// AdminUpdatePasswordLogin PUT /api/admin/config/password-login — 更新密码登录配置。
+func (h *Handler) AdminUpdatePasswordLogin(c *gin.Context) {
+	var req struct {
+		Enabled bool `json:"enabled"`
+	}
+	if err := c.ShouldBindJSON(&req); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid input"})
+		return
+	}
+	value := "false"
+	if req.Enabled {
+		value = "true"
+	}
+	var sc store.SystemConfig
+	if err := h.db.Where("key = ?", "password_login_enabled").First(&sc).Error; err == nil {
+		sc.Value = value
+		h.db.Save(&sc)
+	} else {
+		sc = store.SystemConfig{Key: "password_login_enabled", Value: value}
+		h.db.Create(&sc)
+	}
+	c.JSON(http.StatusOK, gin.H{"ok": true})
+}
diff --git a/backend/internal/api/admin_models.go b/backend/internal/api/admin_models.go
new file mode 100644
index 0000000..588b3e3
--- /dev/null
+++ b/backend/internal/api/admin_models.go
@@ -0,0 +1,291 @@
+package api
+
+import (
+	"encoding/json"
+	"net/http"
+	"strconv"
+
+	"opencatd-open/internal/store"
+
+	"github.com/gin-gonic/gin"
+	"gorm.io/gorm"
+)
+
+// AdminModels GET /api/admin/models — 模型列表（含价格、渠道绑定、定价/禁止状态）。
+func (h *Handler) AdminModels(c *gin.Context) {
+	var ms []store.Model
+	if err := h.db.Order("sort ASC, id ASC").Find(&ms).Error; err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to load models"})
+		return
+	}
+
+	allow, deny := h.modelPolicyConfig()
+
+	out := make([]gin.H, 0, len(ms))
+	for _, m := range ms {
+		var bindings []store.ChannelModelBinding
+		h.db.Preload("Channel").Where("model_id = ?", m.ID).Find(&bindings)
+		chs := make([]gin.H, 0, len(bindings))
+		for _, b := range bindings {
+			if !b.Channel.Enabled {
+				continue
+			}
+			chs = append(chs, gin.H{
+				"id": b.ID, "channel_id": b.ChannelID, "channel_name": b.Channel.Name,
+				"upstream_model": b.UpstreamModel, "weight": b.Weight,
+			})
+		}
+		used := len(chs) > 0
+		needsPricing := used && m.InputPrice == 0 && m.OutputPrice == 0 && m.CacheReadPrice == 0
+		denied := containsStr(deny, m.Name) || (len(allow) > 0 && !containsStr(allow, m.Name))
+		out = append(out, gin.H{
+			"id": m.ID, "name": m.Name, "display_name": m.DisplayName,
+			"input_price": m.InputPrice, "output_price": m.OutputPrice, "cache_read_price": m.CacheReadPrice,
+			"enabled": m.Enabled, "sort": m.Sort, "channels": chs,
+			"used": used, "needs_pricing": needsPricing, "denied": denied,
+		})
+	}
+
+	var orphans []struct {
+		ChannelName   string
+		UpstreamModel string
+		ModelID       uint64
+	}
+	h.db.Raw(`SELECT c.name as channel_name, b.model_id, b.upstream_model
+		FROM channel_model_bindings b
+		LEFT JOIN models m ON m.id = b.model_id
+		LEFT JOIN channels c ON c.id = b.channel_id
+		WHERE m.id IS NULL`).Scan(&orphans)
+	missing := make([]gin.H, 0, len(orphans))
+	for _, o := range orphans {
+		missing = append(missing, gin.H{
+			"channel": o.ChannelName, "model_id": o.ModelID, "upstream_model": o.UpstreamModel,
+		})
+	}
+
+	unpriced := 0
+	{
+		var usedBindings []struct {
+			ModelID uint64
+		}
+		h.db.Model(&store.ChannelModelBinding{}).Distinct("model_id").Scan(&usedBindings)
+		usedIDs := map[uint64]bool{}
+		for _, u := range usedBindings {
+			usedIDs[u.ModelID] = true
+		}
+		for _, m := range ms {
+			if usedIDs[m.ID] && m.InputPrice == 0 && m.OutputPrice == 0 && m.CacheReadPrice == 0 {
+				unpriced++
+			}
+		}
+	}
+
+	c.JSON(http.StatusOK, gin.H{
+		"data": out,
+		"summary": gin.H{
+			"total":        len(ms),
+			"unpriced":     unpriced,
+			"missing":      missing,
+			"denied_count": len(deny),
+		},
+	})
+}
+
+// modelPolicyConfig 读取全局模型允许/禁止列表。
+func (h *Handler) modelPolicyConfig() (allow, deny []string) {
+	var raw string
+	h.db.Model(&store.SystemConfig{}).Where("key = ?", "model_allowlist").Pluck("value", &raw)
+	_ = json.Unmarshal([]byte(raw), &allow)
+	raw = ""
+	h.db.Model(&store.SystemConfig{}).Where("key = ?", "model_denylist").Pluck("value", &raw)
+	_ = json.Unmarshal([]byte(raw), &deny)
+	return
+}
+
+func containsStr(list []string, s string) bool {
+	for _, v := range list {
+		if v == s {
+			return true
+		}
+	}
+	return false
+}
+
+// AdminCreateModel POST /api/admin/models
+func (h *Handler) AdminCreateModel(c *gin.Context) {
+	var req struct {
+		Name           string  `json:"name" binding:"required,min=1,max=128"`
+		DisplayName    string  `json:"display_name"`
+		InputPrice     float64 `json:"input_price"`
+		OutputPrice    float64 `json:"output_price"`
+		CacheReadPrice float64 `json:"cache_read_price"`
+		Sort           int     `json:"sort"`
+		Enabled        *bool   `json:"enabled"`
+	}
+	if err := c.ShouldBindJSON(&req); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid input: " + err.Error()})
+		return
+	}
+	m := store.Model{
+		Name: req.Name, DisplayName: req.DisplayName,
+		InputPrice: req.InputPrice, OutputPrice: req.OutputPrice, CacheReadPrice: req.CacheReadPrice,
+		Sort: req.Sort, Enabled: boolOr(req.Enabled, true),
+	}
+	if err := h.db.Create(&m).Error; err != nil {
+		c.JSON(http.StatusConflict, gin.H{"error": "failed to create model (name may already exist)"})
+		return
+	}
+	c.JSON(http.StatusCreated, gin.H{"id": m.ID, "name": m.Name})
+}
+
+// AdminUpdateModel PUT /api/admin/models/:id — 价格/启停/排序。
+func (h *Handler) AdminUpdateModel(c *gin.Context) {
+	id, err := strconv.ParseUint(c.Param("id"), 10, 64)
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid model id"})
+		return
+	}
+	var req struct {
+		DisplayName    *string  `json:"display_name"`
+		InputPrice     *float64 `json:"input_price"`
+		OutputPrice    *float64 `json:"output_price"`
+		CacheReadPrice *float64 `json:"cache_read_price"`
+		Enabled        *bool    `json:"enabled"`
+		Sort           *int     `json:"sort"`
+	}
+	if err := c.ShouldBindJSON(&req); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid input"})
+		return
+	}
+	var m store.Model
+	if err := h.db.First(&m, id).Error; err != nil {
+		c.JSON(http.StatusNotFound, gin.H{"error": "model not found"})
+		return
+	}
+	updates := map[string]any{}
+	if req.DisplayName != nil {
+		updates["display_name"] = *req.DisplayName
+	}
+	if req.InputPrice != nil {
+		updates["input_price"] = *req.InputPrice
+	}
+	if req.OutputPrice != nil {
+		updates["output_price"] = *req.OutputPrice
+	}
+	if req.CacheReadPrice != nil {
+		updates["cache_read_price"] = *req.CacheReadPrice
+	}
+	if req.Enabled != nil {
+		updates["enabled"] = *req.Enabled
+	}
+	if req.Sort != nil {
+		updates["sort"] = *req.Sort
+	}
+	if len(updates) > 0 {
+		if err := h.db.Model(&m).Updates(updates).Error; err != nil {
+			c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to update model"})
+			return
+		}
+	}
+	c.JSON(http.StatusOK, gin.H{"ok": true})
+}
+
+// AdminDeleteModel DELETE /api/admin/models/:id
+func (h *Handler) AdminDeleteModel(c *gin.Context) {
+	id, err := strconv.ParseUint(c.Param("id"), 10, 64)
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid model id"})
+		return
+	}
+	res := h.db.Delete(&store.Model{}, id)
+	if res.Error != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to delete model"})
+		return
+	}
+	if res.RowsAffected == 0 {
+		c.JSON(http.StatusNotFound, gin.H{"error": "model not found"})
+		return
+	}
+	h.db.Where("model_id = ?", id).Delete(&store.ChannelModelBinding{})
+	c.JSON(http.StatusOK, gin.H{"ok": true})
+}
+
+// AdminDeleteUnusedModels DELETE /api/admin/models/unused — 一键清除未绑定任何渠道的模型。
+func (h *Handler) AdminDeleteUnusedModels(c *gin.Context) {
+	var orphans []store.Model
+	if err := h.db.Where("id NOT IN (SELECT DISTINCT model_id FROM channel_model_bindings)").Find(&orphans).Error; err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to load models"})
+		return
+	}
+	names := make([]string, 0, len(orphans))
+	ids := make([]uint64, 0, len(orphans))
+	for _, m := range orphans {
+		names = append(names, m.Name)
+		ids = append(ids, m.ID)
+	}
+	if len(ids) > 0 {
+		if err := h.db.Delete(&store.Model{}, ids).Error; err != nil {
+			c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to delete models"})
+			return
+		}
+	}
+	c.JSON(http.StatusOK, gin.H{"deleted": names, "count": len(names)})
+}
+
+// AdminCreateModelBinding POST /api/admin/models/:id/bindings
+func (h *Handler) AdminCreateModelBinding(c *gin.Context) {
+	modelID, err := strconv.ParseUint(c.Param("id"), 10, 64)
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid model id"})
+		return
+	}
+	var req struct {
+		ChannelID     uint64 `json:"channel_id" binding:"required"`
+		UpstreamModel string `json:"upstream_model" binding:"required"`
+		Weight        *int   `json:"weight"`
+	}
+	if err := c.ShouldBindJSON(&req); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid input: channel_id and upstream_model required"})
+		return
+	}
+	var m store.Model
+	if err := h.db.First(&m, modelID).Error; err != nil {
+		c.JSON(http.StatusNotFound, gin.H{"error": "model not found"})
+		return
+	}
+	var ch store.Channel
+	if err := h.db.First(&ch, req.ChannelID).Error; err != nil {
+		c.JSON(http.StatusNotFound, gin.H{"error": "channel not found"})
+		return
+	}
+	b := store.ChannelModelBinding{
+		ChannelID: req.ChannelID, ModelID: modelID,
+		UpstreamModel: req.UpstreamModel, Weight: intOr(req.Weight, 1),
+	}
+	if err := h.db.Create(&b).Error; err != nil {
+		c.JSON(http.StatusConflict, gin.H{"error": "binding may already exist"})
+		return
+	}
+	c.JSON(http.StatusCreated, gin.H{"id": b.ID})
+}
+
+// AdminDeleteModelBinding DELETE /api/admin/models/:id/bindings/:bid
+func (h *Handler) AdminDeleteModelBinding(c *gin.Context) {
+	bid, err := strconv.ParseUint(c.Param("bid"), 10, 64)
+	if err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid binding id"})
+		return
+	}
+	res := h.db.Delete(&store.ChannelModelBinding{}, bid)
+	if res.Error != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to delete binding"})
+		return
+	}
+	if res.RowsAffected == 0 {
+		c.JSON(http.StatusNotFound, gin.H{"error": "binding not found"})
+		return
+	}
+	c.JSON(http.StatusOK, gin.H{"ok": true})
+}
+
+var _ = gorm.ErrRecordNotFound
diff --git a/backend/internal/api/api.go b/backend/internal/api/api.go
index 9917986..849efe5 100644
--- a/backend/internal/api/api.go
+++ b/backend/internal/api/api.go
@@ -556,7 +556,7 @@ func (h *Handler) DeleteApiKey(c *gin.Context) {
 	c.JSON(http.StatusOK, gin.H{"message": "deleted"})
 }
 
-// --- Channels ---
+// --- Legacy Channel endpoints (kept for backward compatibility) ---
 
 func (h *Handler) ListChannels(c *gin.Context) {
 	// Support both limit/offset and pageSize/page parameters
@@ -702,7 +702,7 @@ func (h *Handler) DeleteChannel(c *gin.Context) {
 	c.JSON(http.StatusOK, gin.H{"message": "deleted"})
 }
 
-// --- Models ---
+// --- Legacy Model endpoints (kept for backward compatibility) ---
 
 func (h *Handler) ListModels(c *gin.Context) {
 	// Support both limit/offset and pageSize/page parameters
@@ -827,7 +827,7 @@ func (h *Handler) DeleteModel(c *gin.Context) {
 	c.JSON(http.StatusOK, gin.H{"message": "deleted"})
 }
 
-// --- Channel-Model Bindings ---
+// --- Legacy Channel-Model Bindings (kept for backward compatibility) ---
 
 func (h *Handler) BindChannelModels(c *gin.Context) {
 	channelID, err := strconv.ParseUint(c.Param("id"), 10, 64)
diff --git a/backend/internal/channel/channel.go b/backend/internal/channel/channel.go
index 2123258..a7d8bca 100644
--- a/backend/internal/channel/channel.go
+++ b/backend/internal/channel/channel.go
@@ -19,6 +19,9 @@ type Service struct {
 	// Health tracking
 	mu           sync.RWMutex
 	healthStatus map[uint64]*channelHealth
+
+	// Concurrency control per channel
+	sems map[uint64]chan struct{}
 }
 
 type channelHealth struct {
@@ -33,6 +36,7 @@ func NewService(channelDAO *dao.ChannelDAO, modelDAO *dao.ModelDAO) *Service {
 		channelDAO:  channelDAO,
 		modelDAO:    modelDAO,
 		healthStatus: make(map[uint64]*channelHealth),
+		sems:         make(map[uint64]chan struct{}),
 	}
 }
 
@@ -219,3 +223,24 @@ func (s *Service) SelectCandidates(ctx context.Context, modelName string, prefer
 
 	return candidates, nil
 }
+
+// TryAcquire 尝试获取渠道并发槽；渠道满载返回 false（调用方可溢出到其他渠道）。
+// MaxConcurrency<=0 视为不限制。
+func (s *Service) TryAcquire(ch *store.Channel) (func(), bool) {
+	if ch.MaxConcurrency <= 0 {
+		return func() {}, true
+	}
+	s.mu.Lock()
+	sem, ok := s.sems[ch.ID]
+	if !ok {
+		sem = make(chan struct{}, ch.MaxConcurrency)
+		s.sems[ch.ID] = sem
+	}
+	s.mu.Unlock()
+	select {
+	case sem <- struct{}{}:
+		return func() { <-sem }, true
+	default:
+		return nil, false
+	}
+}
diff --git a/backend/internal/channel/health.go b/backend/internal/channel/health.go
index f3a2a19..348acbc 100644
--- a/backend/internal/channel/health.go
+++ b/backend/internal/channel/health.go
@@ -10,19 +10,45 @@ import (
 	"time"
 )
 
+// HealthConfig 健康检查配置
+type HealthConfig struct {
+	Interval          time.Duration // 检查间隔
+	Timeout           time.Duration // 请求超时
+	FailureThreshold  int           // 连续失败次数阈值
+	DegradedCooldown  time.Duration // degraded 冷却时间
+	CooldownCooldown  time.Duration // cooldown 冷却时间
+}
+
+// DefaultHealthConfig 返回默认健康检查配置
+func DefaultHealthConfig() HealthConfig {
+	return HealthConfig{
+		Interval:         5 * time.Minute,
+		Timeout:          10 * time.Second,
+		FailureThreshold: 3,
+		DegradedCooldown: 5 * time.Minute,
+		CooldownCooldown: 15 * time.Minute,
+	}
+}
+
 type HealthChecker struct {
 	channelDAO *dao.ChannelDAO
 	service    *Service
 	client     *http.Client
+	config     HealthConfig
 }
 
-func NewHealthChecker(channelDAO *dao.ChannelDAO, service *Service) *HealthChecker {
+func NewHealthChecker(channelDAO *dao.ChannelDAO, service *Service, config ...HealthConfig) *HealthChecker {
+	cfg := DefaultHealthConfig()
+	if len(config) > 0 {
+		cfg = config[0]
+	}
 	return &HealthChecker{
 		channelDAO: channelDAO,
 		service:    service,
 		client: &http.Client{
-			Timeout: 10 * time.Second,
+			Timeout: cfg.Timeout,
 		},
+		config: cfg,
 	}
 }
 
@@ -91,8 +117,12 @@ func (hc *HealthChecker) CheckAllChannels(ctx context.Context) error {
 }
 
 // StartPeriodicCheck starts periodic health checks
-func (hc *HealthChecker) StartPeriodicCheck(ctx context.Context, interval time.Duration) {
-	ticker := time.NewTicker(interval)
+func (hc *HealthChecker) StartPeriodicCheck(ctx context.Context, interval ...time.Duration) {
+	interval_ := hc.config.Interval
+	if len(interval) > 0 {
+		interval_ = interval[0]
+	}
+	ticker := time.NewTicker(interval_)
 	defer ticker.Stop()
 
 	for {
diff --git a/backend/internal/dao/usage.go b/backend/internal/dao/usage.go
index abbcdb3..0396d72 100644
--- a/backend/internal/dao/usage.go
+++ b/backend/internal/dao/usage.go
@@ -82,10 +82,19 @@ func (d *DailyUsageDAO) GetByDate(ctx context.Context, userID uint64, date strin
 	return &log, nil
 }
 
+// UpsertDailyUsage 按 (user_id, model_id, date) 累加式 upsert：
+// 行不存在则插入；存在则在原值基础上增量累加（不能用 AssignmentColumns 覆盖，
+// 否则多次 flush 会互相清零）。非限定列名在 SQLite/MySQL/PG 的 upsert 语义下都指向目标行。
 func (d *DailyUsageDAO) UpsertDailyUsage(ctx context.Context, log *store.UsageDaily) error {
 	return d.db.WithContext(ctx).Clauses(clause.OnConflict{
-		Columns:   []clause.Column{{Name: "user_id"}, {Name: "model_id"}, {Name: "date"}},
-		DoUpdates: clause.AssignmentColumns([]string{"requests", "input_tokens", "output_tokens", "cache_read_tokens", "cost"}),
+		Columns: []clause.Column{{Name: "user_id"}, {Name: "model_id"}, {Name: "date"}},
+		DoUpdates: clause.Assignments(map[string]interface{}{
+			"requests":          gorm.Expr("requests + ?", log.Requests),
+			"input_tokens":      gorm.Expr("input_tokens + ?", log.InputTokens),
+			"output_tokens":     gorm.Expr("output_tokens + ?", log.OutputTokens),
+			"cache_read_tokens": gorm.Expr("cache_read_tokens + ?", log.CacheReadTokens),
+			"cost":              gorm.Expr("cost + ?", log.Cost),
+		}),
 	}).Create(log).Error
 }
 
diff --git a/backend/internal/proxy/convert/convert.go b/backend/internal/proxy/convert/convert.go
new file mode 100644
index 0000000..dc9cd22
--- /dev/null
+++ b/backend/internal/proxy/convert/convert.go
@@ -0,0 +1,203 @@
+// 三协议互转注册表：OpenAI Chat / OpenAI Responses / Anthropic Messages。
+// 网关以 Chat 形状作为标准中间模型：非跨 chat 的转换经 chat 中转。
+// 请求/响应（非流式）走 JSON 转换；流式走逐行 SSE 转换（stream_transform.go）。
+package convert
+
+import (
+	"bytes"
+	"encoding/json"
+	"fmt"
+)
+
+// 协议标识。
+const (
+	ProtoChat      = "chat"
+	ProtoMessages  = "messages"
+	ProtoResponses = "responses"
+)
+
+// trimBody 去掉首尾空白。部分上游（如 OpenRouter）会在 JSON 前输出空白或
+// SSE 注释行再跟正文，直接 Unmarshal 会失败。
+func trimBody(body []byte) []byte {
+	return bytes.TrimSpace(body)
+}
+
+// CleanJSON 剥离非 JSON 前缀（空白、SSE 注释、`data:` 行）并压缩为标准 JSON。
+// 部分上游（如 OpenRouter）的 non-stream 响应在 JSON 前夹带空白/注释；
+// 原样透传会让客户端解析失败。找不到 JSON 对象时原样返回。
+func CleanJSON(body []byte) []byte {
+	i := bytes.IndexByte(body, '{')
+	if i < 0 {
+		return body
+	}
+	var v any
+	if err := json.Unmarshal(bytes.TrimSpace(body[i:]), &v); err != nil {
+		return body
+	}
+	out, err := json.Marshal(v)
+	if err != nil {
+		return body
+	}
+	return out
+}
+
+// ConvertRequest 转换请求体。from==to 时原样返回。
+func ConvertRequest(body []byte, from, to string) ([]byte, error) {
+	if from == to {
+		return body, nil
+	}
+	body = trimBody(body)
+	switch {
+	case from == ProtoMessages && to == ProtoChat:
+		return messagesToChatReq(body)
+	case from == ProtoChat && to == ProtoMessages:
+		return chatToMessagesReq(body)
+	case from == ProtoResponses && to == ProtoChat:
+		return responsesToChatReq(body)
+	case from == ProtoChat && to == ProtoResponses:
+		return chatToResponsesReq(body)
+	case from == ProtoResponses && to == ProtoMessages:
+		mid, err := responsesToChatReq(body)
+		if err != nil {
+			return nil, err
+		}
+		return chatToMessagesReq(mid)
+	case from == ProtoMessages && to == ProtoResponses:
+		mid, err := messagesToChatReq(body)
+		if err != nil {
+			return nil, err
+		}
+		return chatToResponsesReq(mid)
+	}
+	return nil, fmt.Errorf("unsupported request conversion %s->%s", from, to)
+}
+
+// ConvertResponse 转换响应体（非流式）。from==to 时原样返回。
+func ConvertResponse(body []byte, from, to string) ([]byte, error) {
+	if from == to {
+		return body, nil
+	}
+	body = trimBody(body)
+	switch {
+	case from == ProtoMessages && to == ProtoChat:
+		return messagesToChatResp(body)
+	case from == ProtoChat && to == ProtoMessages:
+		return chatToMessagesResp(body)
+	case from == ProtoResponses && to == ProtoChat:
+		return responsesToChatResp(body)
+	case from == ProtoChat && to == ProtoResponses:
+		return chatToResponsesResp(body)
+	case from == ProtoResponses && to == ProtoMessages:
+		mid, err := responsesToChatResp(body)
+		if err != nil {
+			return nil, err
+		}
+		return chatToMessagesResp(mid)
+	case from == ProtoMessages && to == ProtoResponses:
+		mid, err := messagesToChatResp(body)
+		if err != nil {
+			return nil, err
+		}
+		return chatToResponsesResp(mid)
+	}
+	return nil, fmt.Errorf("unsupported response conversion %s->%s", from, to)
+}
+
+// NewStreamTransformer 构造流式逐行转换器：输入上游 SSE 一行，返回客户端 SSE 行。
+// 返回 nil 表示丢弃该行或无需转换（from==to）。
+func NewStreamTransformer(from, to string) func([]byte) []byte {
+	switch {
+	case from == ProtoMessages && to == ProtoChat:
+		return newMessagesToChat().line
+	case from == ProtoChat && to == ProtoMessages:
+		return newChatToMessages().line
+	case from == ProtoResponses && to == ProtoChat:
+		return newResponsesToChat().line
+	case from == ProtoChat && to == ProtoResponses:
+		return newChatToResponses().line
+	case from == ProtoResponses && to == ProtoMessages:
+		return newResponsesToMessages().line
+	case from == ProtoMessages && to == ProtoResponses:
+		return newMessagesToResponses().line
+	}
+	return nil
+}
+
+// ---------------------------------------------------------------------------
+// 工具函数
+
+// str 返回字符串字段；json.RawMessage 为字符串字面量时去引号。
+func str(raw json.RawMessage) string {
+	if len(raw) == 0 || string(raw) == "null" {
+		return ""
+	}
+	var s string
+	if json.Unmarshal(raw, &s) == nil {
+		return s
+	}
+	// 数组/对象：尝试取 type=text 的 text
+	var arr []map[string]any
+	if json.Unmarshal(raw, &arr) == nil {
+		var parts []string
+		for _, b := range arr {
+			if t, _ := b["type"].(string); t == "text" || t == "input_text" || t == "output_text" {
+				if txt, _ := b["text"].(string); txt != "" {
+					parts = append(parts, txt)
+				}
+			}
+		}
+		return joinNonEmpty(parts, "\n")
+	}
+	return ""
+}
+
+func joinNonEmpty(parts []string, sep string) string {
+	out := ""
+	for _, p := range parts {
+		if p == "" {
+			continue
+		}
+		if out != "" {
+			out += sep
+		}
+		out += p
+	}
+	return out
+}
+
+// rawJSON 安全取字段；不存在或 null 返回 nil。
+func rawJSON(m map[string]json.RawMessage, key string) json.RawMessage {
+	raw, ok := m[key]
+	if !ok || string(raw) == "null" {
+		return nil
+	}
+	return raw
+}
+
+// rawOrObject 把 RawMessage 解为 map；非对象返回空对象。
+func rawOrObject(raw json.RawMessage) any {
+	if len(raw) == 0 || string(raw) == "null" {
+		return map[string]any{}
+	}
+	var m map[string]any
+	if json.Unmarshal(raw, &m) == nil {
+		return m
+	}
+	return map[string]any{}
+}
+
+// intOrNil 取指针值，nil 时返回默认值。
+func intOrNil(p *int, def int) any {
+	if p == nil {
+		return def
+	}
+	return *p
+}
+
+// strField 取 any 中的字符串字段。
+func strField(v any) string {
+	if s, ok := v.(string); ok {
+		return s
+	}
+	return ""
+}
diff --git a/backend/internal/proxy/convert/json_chat.go b/backend/internal/proxy/convert/json_chat.go
new file mode 100644
index 0000000..5edaf6c
--- /dev/null
+++ b/backend/internal/proxy/convert/json_chat.go
@@ -0,0 +1,504 @@
+package convert
+
+import (
+	"encoding/json"
+	"strings"
+)
+
+// ---------------------------------------------------------------------------
+// 请求：Chat → Messages
+
+type chatTool struct {
+	Type     string `json:"type"`
+	Function struct {
+		Name        string          `json:"name"`
+		Description string          `json:"description"`
+		Parameters  json.RawMessage `json:"parameters"`
+	} `json:"function"`
+}
+
+type chatMsg struct {
+	Role       string          `json:"role"`
+	Content    json.RawMessage `json:"content"`
+	ToolCallID string          `json:"tool_call_id"`
+	ToolCalls  []struct {
+		ID       string `json:"id"`
+		Function struct {
+			Name      string `json:"name"`
+			Arguments string `json:"arguments"`
+		} `json:"function"`
+	} `json:"tool_calls"`
+}
+
+type chatReq struct {
+	Model       string      `json:"model"`
+	Messages    []chatMsg   `json:"messages"`
+	Tools       []chatTool  `json:"tools"`
+	Temperature *float64    `json:"temperature"`
+	TopP        *float64    `json:"top_p"`
+	MaxTokens   *int        `json:"max_tokens"`
+	Stop        []string    `json:"stop"`
+	Stream      bool        `json:"stream"`
+}
+
+// chatToMessagesReq 将 OpenAI Chat 请求转为 Anthropic Messages 请求。
+func chatToMessagesReq(body []byte) ([]byte, error) {
+	var req chatReq
+	if err := json.Unmarshal(body, &req); err != nil {
+		return nil, err
+	}
+	out := map[string]any{
+		"model":      req.Model,
+		"max_tokens": intOrNil(req.MaxTokens, 1024), // Anthropic 必填
+	}
+	if req.Stream {
+		out["stream"] = true
+	}
+	if req.Temperature != nil {
+		out["temperature"] = *req.Temperature
+	}
+	if req.TopP != nil {
+		out["top_p"] = *req.TopP
+	}
+	if len(req.Stop) > 0 {
+		out["stop_sequences"] = req.Stop
+	}
+
+	var system []string
+	msgs := make([]any, 0, len(req.Messages))
+	for _, m := range req.Messages {
+		if m.Role == "system" {
+			if s := str(m.Content); s != "" {
+				system = append(system, s)
+			}
+			continue
+		}
+		msgs = append(msgs, chatMsgToAnthropic(m))
+	}
+	if len(system) > 0 {
+		out["system"] = strings.Join(system, "\n")
+	}
+	out["messages"] = msgs
+
+	if len(req.Tools) > 0 {
+		tools := make([]any, 0, len(req.Tools))
+		for _, t := range req.Tools {
+			var params any
+			if len(t.Function.Parameters) > 0 && string(t.Function.Parameters) != "null" {
+				_ = json.Unmarshal(t.Function.Parameters, &params)
+			}
+			tools = append(tools, map[string]any{
+				"name":         t.Function.Name,
+				"description":  t.Function.Description,
+				"input_schema": params,
+			})
+		}
+		out["tools"] = tools
+	}
+	return json.Marshal(out)
+}
+
+// chatMsgToAnthropic 单条消息转 Anthropic 内容。
+func chatMsgToAnthropic(m chatMsg) any {
+	switch m.Role {
+	case "assistant":
+		content := make([]any, 0, 2)
+		if s := str(m.Content); s != "" {
+			content = append(content, map[string]any{"type": "text", "text": s})
+		}
+		for _, tc := range m.ToolCalls {
+			var input any
+			if tc.Function.Arguments != "" {
+				_ = json.Unmarshal([]byte(tc.Function.Arguments), &input)
+			}
+			content = append(content, map[string]any{
+				"type":  "tool_use",
+				"id":    tc.ID,
+				"name":  tc.Function.Name,
+				"input": input,
+			})
+		}
+		return map[string]any{"role": "assistant", "content": content}
+	case "tool":
+		return map[string]any{"role": "user", "content": []any{
+			map[string]any{"type": "tool_result", "tool_use_id": m.ToolCallID, "content": str(m.Content)},
+		}}
+	default: // user
+		var arr []map[string]any
+		if json.Unmarshal(m.Content, &arr) == nil && arr != nil {
+			blocks := make([]any, 0, len(arr))
+			for _, b := range arr {
+				switch b["type"] {
+				case "text", "input_text":
+					if t, _ := b["text"].(string); t != "" {
+						blocks = append(blocks, map[string]any{"type": "text", "text": t})
+					}
+				case "image_url":
+					var url string
+					if iu, ok := b["image_url"].(map[string]any); ok {
+						url, _ = iu["url"].(string)
+					} else if s, ok := b["image_url"].(string); ok {
+						url = s
+					}
+					if url != "" {
+						blocks = append(blocks, anthropicImageBlock(url))
+					}
+				}
+			}
+			if len(blocks) > 0 {
+				return map[string]any{"role": "user", "content": blocks}
+			}
+		}
+		return map[string]any{"role": "user", "content": str(m.Content)}
+	}
+}
+
+// ---------------------------------------------------------------------------
+// 请求：Messages → Chat
+
+type messagesReq struct {
+	Model    string          `json:"model"`
+	System   json.RawMessage `json:"system"`
+	Messages []struct {
+		Role    string          `json:"role"`
+		Content json.RawMessage `json:"content"`
+	} `json:"messages"`
+	Tools []struct {
+		Name        string          `json:"name"`
+		Description string          `json:"description"`
+		InputSchema json.RawMessage `json:"input_schema"`
+	} `json:"tools"`
+	Temperature  *float64 `json:"temperature"`
+	TopP         *float64 `json:"top_p"`
+	MaxTokens    *int     `json:"max_tokens"`
+	StopSequence []string `json:"stop_sequences"`
+	Stream       bool     `json:"stream"`
+}
+
+// messagesToChatReq 将 Anthropic Messages 请求转为 OpenAI Chat 请求。
+func messagesToChatReq(body []byte) ([]byte, error) {
+	var req messagesReq
+	if err := json.Unmarshal(body, &req); err != nil {
+		return nil, err
+	}
+	out := map[string]any{"model": req.Model}
+	if req.Stream {
+		out["stream"] = true
+	}
+	if req.Temperature != nil {
+		out["temperature"] = *req.Temperature
+	}
+	if req.TopP != nil {
+		out["top_p"] = *req.TopP
+	}
+	if req.MaxTokens != nil {
+		out["max_tokens"] = *req.MaxTokens
+	}
+	if len(req.StopSequence) > 0 {
+		out["stop"] = req.StopSequence
+	}
+
+	msgs := make([]any, 0, len(req.Messages)+1)
+	if s := str(req.System); s != "" {
+		msgs = append(msgs, map[string]any{"role": "system", "content": s})
+	}
+	for _, m := range req.Messages {
+		msgs = append(msgs, anthropicMsgToChat(m.Role, m.Content)...)
+	}
+	out["messages"] = msgs
+
+	if len(req.Tools) > 0 {
+		tools := make([]any, 0, len(req.Tools))
+		for _, t := range req.Tools {
+			tools = append(tools, map[string]any{
+				"type": "function",
+				"function": map[string]any{
+					"name":        t.Name,
+					"description": t.Description,
+					"parameters":  rawOrObject(t.InputSchema),
+				},
+			})
+		}
+		out["tools"] = tools
+	}
+	return json.Marshal(out)
+}
+
+// anthropicMsgToChat 将一条 Anthropic 消息拆成 0..N 条 Chat 消息。
+func anthropicMsgToChat(role string, content json.RawMessage) []any {
+	// 块数组优先（tool_use / tool_result 需要分块解析）
+	var blocks []map[string]any
+	if json.Unmarshal(content, &blocks) == nil && blocks != nil {
+		var out []any
+		var toolMsgs []any // tool_result 单独收集，保证排在 assistant(tool_calls) 之后
+		var textParts []string
+		var contentBlocks []any // text / image_url 块，保留原始顺序
+		var toolCalls []any
+		for _, b := range blocks {
+			switch b["type"] {
+			case "text":
+				if t, _ := b["text"].(string); t != "" {
+					textParts = append(textParts, t)
+					contentBlocks = append(contentBlocks, map[string]any{"type": "text", "text": t})
+				}
+			case "image":
+				if cb := chatImageBlock(b); cb != nil {
+					contentBlocks = append(contentBlocks, cb)
+				}
+			case "tool_use":
+				id, _ := b["id"].(string)
+				name, _ := b["name"].(string)
+				args, _ := json.Marshal(b["input"])
+				toolCalls = append(toolCalls, map[string]any{
+					"id":   id,
+					"type": "function",
+					"function": map[string]any{
+						"name":      name,
+						"arguments": string(args),
+					},
+				})
+			case "tool_result":
+				callID, _ := b["tool_use_id"].(string)
+				res := strField(b["content"])
+				toolMsgs = append(toolMsgs, map[string]any{"role": "tool", "tool_call_id": callID, "content": res})
+			}
+		}
+		hasImage := false
+		for _, cb := range contentBlocks {
+			if m, _ := cb.(map[string]any); m["type"] == "image_url" {
+				hasImage = true
+				break
+			}
+		}
+		if hasImage || len(textParts) > 0 || len(toolCalls) > 0 {
+			msg := map[string]any{"role": role}
+			switch {
+			case hasImage:
+				msg["content"] = contentBlocks
+			case len(textParts) > 0:
+				msg["content"] = strings.Join(textParts, "")
+			}
+			if len(toolCalls) > 0 {
+				msg["tool_calls"] = toolCalls
+			}
+			out = append(out, msg)
+		}
+		out = append(out, toolMsgs...)
+		if len(out) > 0 {
+			return out
+		}
+	}
+	// 纯文本
+	if s := str(content); s != "" {
+		return []any{map[string]any{"role": role, "content": s}}
+	}
+	return nil
+}
+
+// ---------------------------------------------------------------------------
+// 响应：Messages → Chat
+
+type messagesResp struct {
+	ID         string `json:"id"`
+	Model      string `json:"model"`
+	Content    []struct {
+		Type  string          `json:"type"`
+		Text  string          `json:"text"`
+		ID    string          `json:"id"`
+		Name  string          `json:"name"`
+		Input json.RawMessage `json:"input"`
+	} `json:"content"`
+	StopReason string `json:"stop_reason"`
+	Usage      struct {
+		InputTokens  int64 `json:"input_tokens"`
+		OutputTokens int64 `json:"output_tokens"`
+	} `json:"usage"`
+}
+
+// messagesToChatResp 将 Anthropic Messages 响应（非流式）转为 Chat 响应。
+func messagesToChatResp(body []byte) ([]byte, error) {
+	var r messagesResp
+	if err := json.Unmarshal(body, &r); err != nil {
+		return nil, err
+	}
+	var text string
+	var toolCalls []any
+	for _, c := range r.Content {
+		switch c.Type {
+		case "text":
+			text += c.Text
+		case "tool_use":
+			args, _ := json.Marshal(c.Input)
+			toolCalls = append(toolCalls, map[string]any{
+				"id":   c.ID,
+				"type": "function",
+				"function": map[string]any{
+					"name":      c.Name,
+					"arguments": string(args),
+				},
+			})
+		}
+	}
+	msg := map[string]any{"role": "assistant", "content": text}
+	if len(toolCalls) > 0 {
+		msg["tool_calls"] = toolCalls
+	}
+	return json.Marshal(map[string]any{
+		"id":      "chatcmpl-" + strings.TrimPrefix(r.ID, "msg_"),
+		"object":  "chat.completion",
+		"model":   r.Model,
+		"created": 0,
+		"choices": []any{map[string]any{
+			"index":         0,
+			"message":       msg,
+			"finish_reason": messagesStopToChat(r.StopReason),
+		}},
+		"usage": map[string]any{
+			"prompt_tokens":     r.Usage.InputTokens,
+			"completion_tokens": r.Usage.OutputTokens,
+			"total_tokens":      r.Usage.InputTokens + r.Usage.OutputTokens,
+		},
+	})
+}
+
+// ---------------------------------------------------------------------------
+// 响应：Chat → Messages
+
+type chatResp struct {
+	ID      string `json:"id"`
+	Model   string `json:"model"`
+	Choices []struct {
+		Message struct {
+			Role      string `json:"role"`
+			Content   string `json:"content"`
+			ToolCalls []struct {
+				ID       string `json:"id"`
+				Function struct {
+					Name      string `json:"name"`
+					Arguments string `json:"arguments"`
+				} `json:"function"`
+			} `json:"tool_calls"`
+		} `json:"message"`
+		FinishReason string `json:"finish_reason"`
+	} `json:"choices"`
+	Usage struct {
+		PromptTokens     int64 `json:"prompt_tokens"`
+		CompletionTokens int64 `json:"completion_tokens"`
+	} `json:"usage"`
+}
+
+// chatToMessagesResp 将 Chat 响应（非流式）转为 Messages 响应。
+func chatToMessagesResp(body []byte) ([]byte, error) {
+	var r chatResp
+	if err := json.Unmarshal(body, &r); err != nil {
+		return nil, err
+	}
+	content := make([]any, 0, 2)
+	var finish = "end_turn"
+	if len(r.Choices) > 0 {
+		msg := r.Choices[0].Message
+		if msg.Content != "" {
+			content = append(content, map[string]any{"type": "text", "text": msg.Content})
+		}
+		for _, tc := range msg.ToolCalls {
+			var input any
+			_ = json.Unmarshal([]byte(tc.Function.Arguments), &input)
+			content = append(content, map[string]any{
+				"type":  "tool_use",
+				"id":    tc.ID,
+				"name":  tc.Function.Name,
+				"input": input,
+			})
+		}
+		finish = chatStopToMessages(r.Choices[0].FinishReason)
+	}
+	return json.Marshal(map[string]any{
+		"id":          "msg_" + strings.TrimPrefix(r.ID, "chatcmpl-"),
+		"type":        "message",
+		"role":        "assistant",
+		"model":       r.Model,
+		"content":     content,
+		"stop_reason": finish,
+		"usage": map[string]any{
+			"input_tokens":  r.Usage.PromptTokens,
+			"output_tokens": r.Usage.CompletionTokens,
+		},
+	})
+}
+
+// ---------------------------------------------------------------------------
+// 辅助
+
+// splitDataURL 解析 data:media_type;base64,data 形式的 URL；非该形式返回 ok=false。
+func splitDataURL(url string) (media, data string, ok bool) {
+	if !strings.HasPrefix(url, "data:") {
+		return "", "", false
+	}
+	i := strings.Index(url, ";base64,")
+	if i < 0 {
+		return "", "", false
+	}
+	return url[len("data:"):i], url[i+len(";base64,"):], true
+}
+
+// chatImageBlock 把 Anthropic image 块转 OpenAI image_url 块。
+// 仅支持 base64 与 url source；其他类型（如 Files API 的 file_id）不支持，跳过。
+func chatImageBlock(b map[string]any) any {
+	src, ok := b["source"].(map[string]any)
+	if !ok {
+		return nil
+	}
+	switch src["type"] {
+	case "base64":
+		media, _ := src["media_type"].(string)
+		data, _ := src["data"].(string)
+		if data == "" {
+			return nil
+		}
+		if media == "" {
+			media = "image/png"
+		}
+		return map[string]any{"type": "image_url", "image_url": map[string]any{"url": "data:" + media + ";base64," + data}}
+	case "url":
+		url, _ := src["url"].(string)
+		if url == "" {
+			return nil
+		}
+		return map[string]any{"type": "image_url", "image_url": map[string]any{"url": url}}
+	}
+	return nil
+}
+
+// anthropicImageBlock 把 OpenAI image_url 的 url 转 Anthropic image 块。
+// data URL → base64 source；http(s) URL → url source。
+func anthropicImageBlock(url string) any {
+	if media, data, ok := splitDataURL(url); ok {
+		if media == "" {
+			media = "image/png"
+		}
+		return map[string]any{"type": "image", "source": map[string]any{"type": "base64", "media_type": media, "data": data}}
+	}
+	return map[string]any{"type": "image", "source": map[string]any{"type": "url", "url": url}}
+}
+
+func messagesStopToChat(s string) string {
+	switch s {
+	case "tool_use":
+		return "tool_calls"
+	case "max_tokens":
+		return "length"
+	default:
+		return "stop"
+	}
+}
+
+func chatStopToMessages(s string) string {
+	switch s {
+	case "tool_calls":
+		return "tool_use"
+	case "length":
+		return "max_tokens"
+	default:
+		return "end_turn"
+	}
+}
diff --git a/backend/internal/proxy/convert/json_responses.go b/backend/internal/proxy/convert/json_responses.go
new file mode 100644
index 0000000..7762e07
--- /dev/null
+++ b/backend/internal/proxy/convert/json_responses.go
@@ -0,0 +1,376 @@
+package convert
+
+import (
+	"encoding/json"
+	"strings"
+)
+
+// ---------------------------------------------------------------------------
+// 请求：Responses → Chat
+
+// responsesToChatReq 将 OpenAI Responses 请求转为 Chat 请求。
+func responsesToChatReq(body []byte) ([]byte, error) {
+	var m map[string]json.RawMessage
+	if err := json.Unmarshal(body, &m); err != nil {
+		return nil, err
+	}
+	out := map[string]any{"model": str(rawJSON(m, "model"))}
+	if v, ok := m["stream"]; ok && string(v) == "true" {
+		out["stream"] = true
+	}
+	if v, ok := m["temperature"]; ok {
+		out["temperature"] = v
+	}
+	if v, ok := m["top_p"]; ok {
+		out["top_p"] = v
+	}
+	if v, ok := m["max_output_tokens"]; ok {
+		out["max_tokens"] = v
+	}
+
+	var msgs []any
+	if ins := str(rawJSON(m, "instructions")); ins != "" {
+		msgs = append(msgs, map[string]any{"role": "system", "content": ins})
+	}
+	msgs = append(msgs, responsesInputToChat(rawJSON(m, "input"))...)
+	out["messages"] = msgs
+
+	if raw := rawJSON(m, "tools"); raw != nil {
+		var tools []map[string]any
+		if json.Unmarshal(raw, &tools) == nil {
+			chatTools := make([]any, 0, len(tools))
+			for _, t := range tools {
+				chatTools = append(chatTools, map[string]any{
+					"type": "function",
+					"function": map[string]any{
+						"name":        t["name"],
+						"description": t["description"],
+						"parameters":  t["parameters"],
+					},
+				})
+			}
+			out["tools"] = chatTools
+		}
+	}
+	return json.Marshal(out)
+}
+
+// responsesInputToChat 把 Responses input 转成 Chat messages。
+// input 支持字符串或条目数组（message / function_call / function_call_output）。
+func responsesInputToChat(raw json.RawMessage) []any {
+	if len(raw) == 0 || string(raw) == "null" {
+		return nil
+	}
+	// 字符串输入
+	if s := str(raw); s != "" {
+		return []any{map[string]any{"role": "user", "content": s}}
+	}
+	var items []map[string]any
+	if err := json.Unmarshal(raw, &items); err != nil || items == nil {
+		return nil
+	}
+	var out []any
+	for _, item := range items {
+		switch item["type"] {
+		case "function_call":
+			out = append(out, map[string]any{
+				"role": "assistant",
+				"content": "",
+				"tool_calls": []any{map[string]any{
+					"id":   strField(item["call_id"]),
+					"type": "function",
+					"function": map[string]any{
+						"name":      strField(item["name"]),
+						"arguments": strField(item["arguments"]),
+					},
+				}},
+			})
+		case "function_call_output":
+			out = append(out, map[string]any{
+				"role":         "tool",
+				"tool_call_id": strField(item["call_id"]),
+				"content":      strField(item["output"]),
+			})
+		default: // message 条目
+			role, _ := item["role"].(string)
+			if role == "" {
+				role = "user"
+			}
+			if content, ok := item["content"].(string); ok {
+				out = append(out, map[string]any{"role": role, "content": content})
+			} else if blocks, ok := item["content"].([]any); ok {
+				var text []string
+				var contentBlocks []any
+				for _, b := range blocks {
+					bm, ok := b.(map[string]any)
+					if !ok {
+						continue
+					}
+					switch bm["type"] {
+					case "input_text", "text":
+						if t, _ := bm["text"].(string); t != "" {
+							text = append(text, t)
+							contentBlocks = append(contentBlocks, map[string]any{"type": "text", "text": t})
+						}
+					case "input_image":
+						var url string
+						if s, ok := bm["image_url"].(string); ok {
+							url = s
+						} else if m, ok := bm["image_url"].(map[string]any); ok {
+							url, _ = m["url"].(string)
+						}
+						if url != "" {
+							contentBlocks = append(contentBlocks, map[string]any{"type": "image_url", "image_url": map[string]any{"url": url}})
+						}
+					}
+				}
+				hasImage := false
+				for _, cb := range contentBlocks {
+					if m, _ := cb.(map[string]any); m["type"] == "image_url" {
+						hasImage = true
+						break
+					}
+				}
+				if hasImage {
+					out = append(out, map[string]any{"role": role, "content": contentBlocks})
+				} else {
+					out = append(out, map[string]any{"role": role, "content": strings.Join(text, "")})
+				}
+			}
+		}
+	}
+	return out
+}
+
+// chatContentToResponsesBlocks 把 Chat 用户消息 content 转 Responses input 块数组（input_text / input_image）。
+func chatContentToResponsesBlocks(content json.RawMessage) []any {
+	// 纯字符串 → 单个 input_text
+	var s string
+	if json.Unmarshal(content, &s) == nil && s != "" {
+		return []any{map[string]any{"type": "input_text", "text": s}}
+	}
+	// 数组 → 按块转换（text / image_url）
+	var arr []map[string]any
+	if json.Unmarshal(content, &arr) == nil && arr != nil {
+		var out []any
+		for _, b := range arr {
+			switch b["type"] {
+			case "text", "input_text":
+				if t, _ := b["text"].(string); t != "" {
+					out = append(out, map[string]any{"type": "input_text", "text": t})
+				}
+			case "image_url":
+				var url string
+				if iu, ok := b["image_url"].(map[string]any); ok {
+					url, _ = iu["url"].(string)
+				} else if s, ok := b["image_url"].(string); ok {
+					url = s
+				}
+				if url != "" {
+					out = append(out, map[string]any{"type": "input_image", "image_url": url})
+				}
+			}
+		}
+		return out
+	}
+	return nil
+}
+
+// ---------------------------------------------------------------------------
+// 请求：Chat → Responses
+
+// chatToResponsesReq 将 Chat 请求转为 Responses 请求。
+func chatToResponsesReq(body []byte) ([]byte, error) {
+	var req chatReq
+	if err := json.Unmarshal(body, &req); err != nil {
+		return nil, err
+	}
+	out := map[string]any{"model": req.Model}
+	if req.Stream {
+		out["stream"] = true
+	}
+	if req.Temperature != nil {
+		out["temperature"] = *req.Temperature
+	}
+	if req.TopP != nil {
+		out["top_p"] = *req.TopP
+	}
+	if req.MaxTokens != nil {
+		out["max_output_tokens"] = *req.MaxTokens
+	}
+
+	var system []string
+	var input []any
+	for _, m := range req.Messages {
+		if m.Role == "system" {
+			if s := str(m.Content); s != "" {
+				system = append(system, s)
+			}
+			continue
+		}
+		switch m.Role {
+		case "tool":
+			input = append(input, map[string]any{
+				"type":   "function_call_output",
+				"call_id": m.ToolCallID,
+				"output": str(m.Content),
+			})
+		case "assistant":
+			if len(m.ToolCalls) > 0 {
+				for _, tc := range m.ToolCalls {
+					input = append(input, map[string]any{
+						"type":      "function_call",
+						"call_id":   tc.ID,
+						"name":      tc.Function.Name,
+						"arguments": tc.Function.Arguments,
+					})
+				}
+			} else if s := str(m.Content); s != "" {
+				input = append(input, map[string]any{"type": "message", "role": "assistant", "content": []any{
+					map[string]any{"type": "input_text", "text": s},
+				}})
+			}
+		default:
+			if blocks := chatContentToResponsesBlocks(m.Content); len(blocks) > 0 {
+				input = append(input, map[string]any{"type": "message", "role": "user", "content": blocks})
+			}
+		}
+	}
+	if len(system) > 0 {
+		out["instructions"] = strings.Join(system, "\n")
+	}
+	// input 必须是数组：部分上游只接受数组，单对象会被拒（400 Mismatch type）。
+	out["input"] = input
+
+	if len(req.Tools) > 0 {
+		tools := make([]any, 0, len(req.Tools))
+		for _, t := range req.Tools {
+			tools = append(tools, map[string]any{
+				"type":        "function",
+				"name":        t.Function.Name,
+				"description": t.Function.Description,
+				"parameters":  rawOrObject(t.Function.Parameters),
+			})
+		}
+		out["tools"] = tools
+	}
+	return json.Marshal(out)
+}
+
+// ---------------------------------------------------------------------------
+// 响应：Responses → Chat
+
+// responsesToChatResp 将 Responses 响应（非流式）转为 Chat 响应。
+func responsesToChatResp(body []byte) ([]byte, error) {
+	var m map[string]json.RawMessage
+	if err := json.Unmarshal(body, &m); err != nil {
+		return nil, err
+	}
+	var text string
+	var toolCalls []any
+	if raw := rawJSON(m, "output"); raw != nil {
+		var outputs []map[string]any
+		if json.Unmarshal(raw, &outputs) == nil {
+			for _, o := range outputs {
+				switch o["type"] {
+				case "message":
+					if content, ok := o["content"].([]any); ok {
+						for _, c := range content {
+							if cm, ok := c.(map[string]any); ok {
+								if t, _ := cm["text"].(string); t != "" {
+									text += t
+								}
+							}
+						}
+					}
+				case "function_call":
+					toolCalls = append(toolCalls, map[string]any{
+						"id":   strField(o["call_id"]),
+						"type": "function",
+						"function": map[string]any{
+							"name":      strField(o["name"]),
+							"arguments": strField(o["arguments"]),
+						},
+					})
+				}
+			}
+		}
+	}
+	msg := map[string]any{"role": "assistant", "content": text}
+	if len(toolCalls) > 0 {
+		msg["tool_calls"] = toolCalls
+	}
+	finish := "stop"
+	switch {
+	case string(rawJSON(m, "status")) == `"incomplete"`:
+		finish = "length" // 截断优先，客户端可据此区分
+	case len(toolCalls) > 0:
+		finish = "tool_calls"
+	}
+	var prompt, completion int64
+	if u := rawJSON(m, "usage"); u != nil {
+		var us struct {
+			InputTokens  int64 `json:"input_tokens"`
+			OutputTokens int64 `json:"output_tokens"`
+		}
+		_ = json.Unmarshal(u, &us)
+		prompt, completion = us.InputTokens, us.OutputTokens
+	}
+	return json.Marshal(map[string]any{
+		"id":      "chatcmpl-" + strings.TrimPrefix(str(rawJSON(m, "id")), "resp_"),
+		"object":  "chat.completion",
+		"model":   str(rawJSON(m, "model")),
+		"choices": []any{map[string]any{"index": 0, "message": msg, "finish_reason": finish}},
+		"usage": map[string]any{
+			"prompt_tokens":     prompt,
+			"completion_tokens": completion,
+			"total_tokens":      prompt + completion,
+		},
+	})
+}
+
+// ---------------------------------------------------------------------------
+// 响应：Chat → Responses
+
+// chatToResponsesResp 将 Chat 响应（非流式）转为 Responses 响应。
+func chatToResponsesResp(body []byte) ([]byte, error) {
+	var r chatResp
+	if err := json.Unmarshal(body, &r); err != nil {
+		return nil, err
+	}
+	output := make([]any, 0, 2)
+	var finish = "completed"
+	if len(r.Choices) > 0 {
+		msg := r.Choices[0].Message
+		if msg.Content != "" {
+			output = append(output, map[string]any{
+				"type":    "message",
+				"role":    "assistant",
+				"content": []any{map[string]any{"type": "output_text", "text": msg.Content}},
+			})
+		}
+		for _, tc := range msg.ToolCalls {
+			output = append(output, map[string]any{
+				"type":      "function_call",
+				"call_id":   tc.ID,
+				"name":      tc.Function.Name,
+				"arguments": tc.Function.Arguments,
+			})
+		}
+		if r.Choices[0].FinishReason == "length" {
+			finish = "incomplete"
+		}
+	}
+	return json.Marshal(map[string]any{
+		"id":     "resp_" + strings.TrimPrefix(r.ID, "chatcmpl-"),
+		"object": "response",
+		"model":  r.Model,
+		"status": finish,
+		"output": output,
+		"usage": map[string]any{
+			"input_tokens":  r.Usage.PromptTokens,
+			"output_tokens": r.Usage.CompletionTokens,
+			"total_tokens":  r.Usage.PromptTokens + r.Usage.CompletionTokens,
+		},
+	})
+}
diff --git a/backend/internal/proxy/convert/stream_transform.go b/backend/internal/proxy/convert/stream_transform.go
new file mode 100644
index 0000000..fdd5c08
--- /dev/null
+++ b/backend/internal/proxy/convert/stream_transform.go
@@ -0,0 +1,642 @@
+package convert
+
+import (
+	"encoding/json"
+	"strings"
+)
+
+// sseState 记录上一行 event 名。
+type sseState struct {
+	event string
+}
+
+// parseLine 解析一行 SSE；返回是否 data 行及其内容、是否 [DONE]。
+// data: 后可跟空格（标准）或紧贴 JSON（部分上游会省略空格）。
+func (s *sseState) parseLine(line []byte) (isData bool, data string, done bool) {
+	strLine := strings.TrimRight(string(line), "\r\n")
+	switch {
+	case strings.HasPrefix(strLine, "event: "):
+		s.event = strings.TrimSpace(strings.TrimPrefix(strLine, "event: "))
+		return false, "", false
+	case strLine == "data: [DONE]" || strLine == "data:[DONE]":
+		return true, "[DONE]", true
+	case strings.HasPrefix(strLine, "data:"):
+		return true, strings.TrimLeft(strings.TrimPrefix(strLine, "data:"), " "), false
+	default:
+		return false, "", false
+	}
+}
+
+func eventData(line string) map[string]any {
+	var m map[string]any
+	_ = json.Unmarshal([]byte(line), &m)
+	return m
+}
+
+func dataLine(obj any) []byte {
+	b, _ := json.Marshal(obj)
+	return append(append([]byte("data: "), b...), '\n', '\n')
+}
+
+func eventLine(name string, obj any) []byte {
+	b, _ := json.Marshal(obj)
+	out := append([]byte("event: "+name+"\ndata: "), b...)
+	return append(out, '\n', '\n')
+}
+
+// joinLines 拼接多条 SSE 行。
+func joinLines(lines [][]byte) []byte {
+	var s []string
+	for _, l := range lines {
+		s = append(s, string(l))
+	}
+	return []byte(strings.Join(s, ""))
+}
+
+// ---------------------------------------------------------------------------
+// Messages → Chat
+
+type messagesToChat struct {
+	sseState
+	id, model string
+	toolIdx   map[int]int // messages content block index → chat tool_calls index（顺序编号，避开文本块）
+	nextTool  int
+}
+
+func newMessagesToChat() *messagesToChat { return &messagesToChat{toolIdx: map[int]int{}} }
+
+func (t *messagesToChat) line(line []byte) []byte {
+	isData, data, done := t.parseLine(line)
+	if !isData {
+		return nil
+	}
+	if done {
+		return []byte("data: [DONE]\n\n")
+	}
+	m := eventData(data)
+	evt, _ := m["type"].(string)
+	switch evt {
+	case "message_start":
+		msg, _ := m["message"].(map[string]any)
+		t.id, _ = msg["id"].(string)
+		t.model, _ = msg["model"].(string)
+		return dataLine(map[string]any{
+			"id": "chatcmpl-" + strings.TrimPrefix(t.id, "msg_"), "object": "chat.completion.chunk", "model": t.model,
+			"choices": []any{map[string]any{"index": 0, "delta": map[string]any{"role": "assistant"}, "finish_reason": nil}},
+		})
+	case "content_block_start":
+		cb, _ := m["content_block"].(map[string]any)
+		if cb == nil || cb["type"] != "tool_use" {
+			return nil
+		}
+		blockIdx, _ := m["index"].(float64)
+		tool := t.nextTool
+		t.nextTool++
+		t.toolIdx[int(blockIdx)] = tool
+		toolID, _ := cb["id"].(string)
+		name, _ := cb["name"].(string)
+		return dataLine(map[string]any{
+			"id": "chatcmpl-" + strings.TrimPrefix(t.id, "msg_"), "object": "chat.completion.chunk", "model": t.model,
+			"choices": []any{map[string]any{"index": 0, "delta": map[string]any{
+				"tool_calls": []any{map[string]any{"index": tool, "id": toolID, "type": "function", "function": map[string]any{"name": name, "arguments": ""}}},
+			}, "finish_reason": nil}},
+		})
+	case "content_block_delta":
+		delta, _ := m["delta"].(map[string]any)
+		deltaType, _ := delta["type"].(string)
+		if deltaType == "input_json_delta" {
+			blockIdx, _ := m["index"].(float64)
+			tool, ok := t.toolIdx[int(blockIdx)]
+			if !ok {
+				return nil
+			}
+			partial, _ := delta["partial_json"].(string)
+			if partial == "" {
+				return nil
+			}
+			return dataLine(map[string]any{
+				"id": "chatcmpl-" + strings.TrimPrefix(t.id, "msg_"), "object": "chat.completion.chunk", "model": t.model,
+				"choices": []any{map[string]any{"index": 0, "delta": map[string]any{
+					"tool_calls": []any{map[string]any{"index": tool, "function": map[string]any{"arguments": partial}}},
+				}, "finish_reason": nil}},
+			})
+		}
+		text, _ := delta["text"].(string)
+		if text == "" {
+			return nil
+		}
+		return dataLine(map[string]any{
+			"id": "chatcmpl-" + strings.TrimPrefix(t.id, "msg_"), "object": "chat.completion.chunk", "model": t.model,
+			"choices": []any{map[string]any{"index": 0, "delta": map[string]any{"content": text}, "finish_reason": nil}},
+		})
+	case "message_delta":
+		delta, _ := m["delta"].(map[string]any)
+		stop, _ := delta["stop_reason"].(string)
+		var out [][]byte
+		out = append(out, dataLine(map[string]any{
+			"id": "chatcmpl-" + strings.TrimPrefix(t.id, "msg_"), "object": "chat.completion.chunk", "model": t.model,
+			"choices": []any{map[string]any{"index": 0, "delta": map[string]any{}, "finish_reason": messagesStopToChat(stop)}},
+		}))
+		if u, ok := m["usage"]; ok {
+			out = append(out, dataLine(map[string]any{
+				"id": "chatcmpl-" + strings.TrimPrefix(t.id, "msg_"), "object": "chat.completion.chunk", "model": t.model,
+				"choices": []any{}, "usage": u,
+			}))
+		}
+		return joinLines(out)
+	case "message_stop":
+		return []byte("data: [DONE]\n\n")
+	}
+	return nil
+}
+
+// ---------------------------------------------------------------------------
+// Chat → Messages
+
+type chatToMessages struct {
+	sseState
+	started    bool    // message_start 已发出
+	nextIndex  int     // 下一个 content block index（顺序分配）
+	textIndex  int     // 文本块 index；-1 = 未开始
+	toolIdx    map[int]int // chat delta.tool_calls[].index → messages block index
+	openBlocks []int   // 已开始未停止的 block index，按开始顺序
+	model      string
+	stopReason string
+	usage      any
+}
+
+func newChatToMessages() *chatToMessages {
+	return &chatToMessages{textIndex: -1, toolIdx: map[int]int{}}
+}
+
+func (t *chatToMessages) line(line []byte) []byte {
+	isData, data, done := t.parseLine(line)
+	if !isData {
+		return nil
+	}
+	if done {
+		// 汇聚最终：先对每个已开始未停止的块发 content_block_stop，再 message_delta + message_stop
+		var out [][]byte
+		for _, idx := range t.openBlocks {
+			out = append(out, eventLine("content_block_stop", map[string]any{"type": "content_block_stop", "index": idx}))
+		}
+		md := map[string]any{"type": "message_delta", "delta": map[string]any{
+			"stop_reason": stopReasonOrEnd(t.stopReason), "stop_sequence": nil,
+		}}
+		if t.usage != nil {
+			md["usage"] = t.usage
+		}
+		out = append(out, eventLine("message_delta", md))
+		out = append(out, eventLine("message_stop", map[string]any{"type": "message_stop"}))
+		return joinLines(out)
+	}
+	m := eventData(data)
+	// chat 块：delta / finish_reason 在 choices[0] 内
+	delta := map[string]any{}
+	if choices, ok := m["choices"].([]any); ok && len(choices) > 0 {
+		if c0, ok := choices[0].(map[string]any); ok {
+			if d, ok := c0["delta"].(map[string]any); ok {
+				delta = d
+			}
+			if fr, _ := c0["finish_reason"].(string); fr != "" {
+				t.stopReason = fr
+			}
+		}
+	}
+	if t.model == "" {
+		t.model, _ = m["model"].(string)
+	}
+	id, _ := m["id"].(string)
+
+	var out [][]byte
+	// message_start 只在实际有内容（文本或工具）时发出，避免 reasoning_content 块
+	//（带 role 无 content）提前开出一个空文本块。
+	ensureStarted := func() {
+		if t.started {
+			return
+		}
+		t.started = true
+		out = append(out, eventLine("message_start", map[string]any{
+			"type": "message_start",
+			"message": map[string]any{
+				"id": "msg_" + strings.TrimPrefix(id, "chatcmpl-"), "type": "message", "role": "assistant",
+				"model": t.model, "content": []any{}, "usage": map[string]any{"input_tokens": 0, "output_tokens": 0},
+			},
+		}))
+	}
+
+	// 文本：delta.content（string；兼容 {type:text,text} 数组）
+	if content := deltaText(delta); content != "" {
+		if t.textIndex < 0 {
+			t.textIndex = t.nextIndex
+			t.nextIndex++
+			ensureStarted()
+			out = append(out, eventLine("content_block_start", map[string]any{
+				"type": "content_block_start", "index": t.textIndex, "content_block": map[string]any{"type": "text", "text": ""},
+			}))
+			t.openBlocks = append(t.openBlocks, t.textIndex)
+		}
+		out = append(out, eventLine("content_block_delta", map[string]any{
+			"type": "content_block_delta", "index": t.textIndex, "delta": map[string]any{"type": "text_delta", "text": content},
+		}))
+	}
+
+	// 工具调用：delta.tool_calls（并行调用各 index 独立成块；arguments 支持整段/分段两种流式）
+	if tcs, ok := delta["tool_calls"].([]any); ok {
+		for _, tc := range tcs {
+			call, ok := tc.(map[string]any)
+			if !ok {
+				continue
+			}
+			idx, _ := call["index"].(float64)
+			tcIdx := int(idx)
+			fn, _ := call["function"].(map[string]any)
+			name, _ := fn["name"].(string)
+			args, _ := fn["arguments"].(string)
+			blockIdx, seen := t.toolIdx[tcIdx]
+			if !seen {
+				blockIdx = t.nextIndex
+				t.nextIndex++
+				t.toolIdx[tcIdx] = blockIdx
+				toolID, _ := call["id"].(string)
+				ensureStarted()
+				out = append(out, eventLine("content_block_start", map[string]any{
+					"type": "content_block_start", "index": blockIdx, "content_block": map[string]any{
+						"type": "tool_use", "id": toolID, "name": name, "input": map[string]any{},
+					},
+				}))
+				t.openBlocks = append(t.openBlocks, blockIdx)
+			}
+			if args != "" {
+				out = append(out, eventLine("content_block_delta", map[string]any{
+					"type": "content_block_delta", "index": blockIdx, "delta": map[string]any{"type": "input_json_delta", "partial_json": args},
+				}))
+			}
+		}
+	}
+
+	if u, ok := m["usage"]; ok {
+		t.usage = u
+	}
+	return joinLines(out)
+}
+
+// deltaText 取 chat delta.content 文本（string 或 [{type:text,text}] 数组拼接）。
+func deltaText(delta map[string]any) string {
+	if s, ok := delta["content"].(string); ok {
+		return s
+	}
+	if arr, ok := delta["content"].([]any); ok {
+		var parts []string
+		for _, b := range arr {
+			if bm, ok := b.(map[string]any); ok {
+				if t, _ := bm["text"].(string); t != "" {
+					parts = append(parts, t)
+				}
+			}
+		}
+		return strings.Join(parts, "")
+	}
+	return ""
+}
+
+func stopReasonOrEnd(s string) string {
+	if s == "" {
+		return "end_turn"
+	}
+	return chatStopToMessages(s)
+}
+
+// ---------------------------------------------------------------------------
+// Responses → Messages
+
+type responsesToMessages struct {
+	sseState
+	started    bool
+	model      string
+	usage      any
+	nextIndex  int            // 下一个 content block index（顺序分配）
+	textIndex  int            // 文本块 index；-1 = 未开始
+	toolIdx    map[string]int // function_call item_id → messages block index
+	openBlocks []int          // 已开始未停止的 block index，按开始顺序
+	anyTool    bool
+}
+
+func newResponsesToMessages() *responsesToMessages {
+	return &responsesToMessages{textIndex: -1, toolIdx: map[string]int{}}
+}
+
+func (t *responsesToMessages) line(line []byte) []byte {
+	isData, data, done := t.parseLine(line)
+	if !isData || done {
+		return nil
+	}
+	m := eventData(data)
+	evt, _ := m["type"].(string)
+	if resp, ok := m["response"].(map[string]any); ok {
+		if t.model == "" {
+			t.model, _ = resp["model"].(string)
+		}
+		if u, ok := resp["usage"]; ok {
+			t.usage = u
+		}
+	}
+	var out [][]byte
+	// message_start 只在 response.created 时发出；文本/工具块在对应事件到达时再开，
+	// 避免纯函数调用响应提前开出一个空文本块。
+	ensureStarted := func() {
+		if t.started {
+			return
+		}
+		t.started = true
+		rid := ""
+		if resp, ok := m["response"].(map[string]any); ok {
+			rid, _ = resp["id"].(string)
+		}
+		out = append(out, eventLine("message_start", map[string]any{
+			"type": "message_start",
+			"message": map[string]any{
+				"id": "msg_" + strings.TrimPrefix(rid, "resp_"), "type": "message", "role": "assistant",
+				"model": t.model, "content": []any{},
+			},
+		}))
+	}
+	switch evt {
+	case "response.created":
+		ensureStarted()
+	case "response.output_text.delta":
+		delta, _ := m["delta"].(string)
+		if delta == "" {
+			return nil
+		}
+		if t.textIndex < 0 {
+			t.textIndex = t.nextIndex
+			t.nextIndex++
+			ensureStarted()
+			out = append(out, eventLine("content_block_start", map[string]any{
+				"type": "content_block_start", "index": t.textIndex, "content_block": map[string]any{"type": "text", "text": ""},
+			}))
+			t.openBlocks = append(t.openBlocks, t.textIndex)
+		}
+		out = append(out, eventLine("content_block_delta", map[string]any{
+			"type": "content_block_delta", "index": t.textIndex, "delta": map[string]any{"type": "text_delta", "text": delta},
+		}))
+	case "response.output_item.added":
+		item, _ := m["item"].(map[string]any)
+		if item == nil || item["type"] != "function_call" {
+			return nil
+		}
+		blockIdx := t.nextIndex
+		t.nextIndex++
+		t.anyTool = true
+		itemID, _ := item["id"].(string)
+		t.toolIdx[itemID] = blockIdx
+		toolUseID, _ := item["call_id"].(string)
+		if toolUseID == "" {
+			toolUseID = itemID
+		}
+		name, _ := item["name"].(string)
+		ensureStarted()
+		out = append(out, eventLine("content_block_start", map[string]any{
+			"type": "content_block_start", "index": blockIdx, "content_block": map[string]any{
+				"type": "tool_use", "id": toolUseID, "name": name, "input": map[string]any{},
+			},
+		}))
+		t.openBlocks = append(t.openBlocks, blockIdx)
+	case "response.function_call_arguments.delta":
+		itemID, _ := m["item_id"].(string)
+		blockIdx, ok := t.toolIdx[itemID]
+		if !ok {
+			return nil
+		}
+		delta, _ := m["delta"].(string)
+		if delta == "" {
+			return nil
+		}
+		out = append(out, eventLine("content_block_delta", map[string]any{
+			"type": "content_block_delta", "index": blockIdx, "delta": map[string]any{"type": "input_json_delta", "partial_json": delta},
+		}))
+	case "response.completed":
+		for _, idx := range t.openBlocks {
+			out = append(out, eventLine("content_block_stop", map[string]any{"type": "content_block_stop", "index": idx}))
+		}
+		stop := "end_turn"
+		if t.anyTool {
+			stop = "tool_use"
+		}
+		md := map[string]any{"type": "message_delta", "delta": map[string]any{"stop_reason": stop, "stop_sequence": nil}}
+		if t.usage != nil {
+			md["usage"] = t.usage
+		}
+		out = append(out, eventLine("message_delta", md))
+		out = append(out, eventLine("message_stop", map[string]any{"type": "message_stop"}))
+	}
+	return joinLines(out)
+}
+
+// ---------------------------------------------------------------------------
+// Messages → Responses
+
+type messagesToResponses struct {
+	sseState
+	model string
+	usage any
+	done  bool
+}
+
+func newMessagesToResponses() *messagesToResponses { return &messagesToResponses{} }
+
+func (t *messagesToResponses) line(line []byte) []byte {
+	isData, data, done := t.parseLine(line)
+	if !isData || done {
+		return nil
+	}
+	m := eventData(data)
+	evt, _ := m["type"].(string)
+	if msg, ok := m["message"].(map[string]any); ok {
+		if t.model == "" {
+			t.model, _ = msg["model"].(string)
+		}
+		if u, ok := msg["usage"]; ok {
+			t.usage = u
+		}
+	}
+	if u, ok := m["usage"]; ok {
+		t.usage = u
+	}
+	var out [][]byte
+	switch evt {
+	case "message_start":
+		id, _ := m["message"].(map[string]any)
+		rid := ""
+		if id != nil {
+			rid, _ = id["id"].(string)
+		}
+		out = append(out, eventLine("response.created", map[string]any{
+			"type": "response.created",
+			"response": map[string]any{
+				"id": "resp_" + strings.TrimPrefix(rid, "msg_"), "object": "response", "model": t.model, "status": "in_progress",
+			},
+		}))
+	case "content_block_delta":
+		delta, _ := m["delta"].(map[string]any)
+		text, _ := delta["text"].(string)
+		if text != "" {
+			out = append(out, eventLine("response.output_text.delta", map[string]any{
+				"type": "response.output_text.delta", "delta": text, "item_id": "msg_1", "output_index": 0, "content_index": 0,
+			}))
+		}
+	case "message_stop":
+		if !t.done {
+			t.done = true
+			out = append(out, eventLine("response.completed", map[string]any{
+				"type": "response.completed",
+				"response": map[string]any{
+					"id": "resp_stream", "object": "response", "model": t.model, "status": "completed", "usage": t.usage,
+				},
+			}))
+		}
+	}
+	return joinLines(out)
+}
+
+// ---------------------------------------------------------------------------
+// Responses → Chat
+
+type responsesToChat struct {
+	sseState
+	id, model string
+}
+
+func newResponsesToChat() *responsesToChat { return &responsesToChat{} }
+
+func (t *responsesToChat) line(line []byte) []byte {
+	isData, data, done := t.parseLine(line)
+	if !isData {
+		return nil
+	}
+	if done {
+		return nil
+	}
+	m := eventData(data)
+	evt, _ := m["type"].(string)
+	if resp, ok := m["response"].(map[string]any); ok {
+		if t.model == "" {
+			t.model, _ = resp["model"].(string)
+		}
+		if t.id == "" {
+			t.id, _ = resp["id"].(string)
+		}
+	}
+	var out [][]byte
+	switch evt {
+	case "response.created":
+		out = append(out, dataLine(map[string]any{
+			"id": "chatcmpl-" + strings.TrimPrefix(t.id, "resp_"), "object": "chat.completion.chunk", "model": t.model,
+			"choices": []any{map[string]any{"index": 0, "delta": map[string]any{"role": "assistant"}, "finish_reason": nil}},
+		}))
+	case "response.output_text.delta":
+		delta, _ := m["delta"].(string)
+		if delta != "" {
+			out = append(out, dataLine(map[string]any{
+				"id": "chatcmpl-" + strings.TrimPrefix(t.id, "resp_"), "object": "chat.completion.chunk", "model": t.model,
+				"choices": []any{map[string]any{"index": 0, "delta": map[string]any{"content": delta}, "finish_reason": nil}},
+			}))
+		}
+	case "response.completed":
+		out = append(out, dataLine(map[string]any{
+			"id": "chatcmpl-" + strings.TrimPrefix(t.id, "resp_"), "object": "chat.completion.chunk", "model": t.model,
+			"choices": []any{map[string]any{"index": 0, "delta": map[string]any{}, "finish_reason": "stop"}},
+		}))
+		if u, ok := m["response"].(map[string]any); ok {
+			if usage, ok := u["usage"]; ok {
+				out = append(out, dataLine(map[string]any{
+					"id": "chatcmpl-" + strings.TrimPrefix(t.id, "resp_"), "object": "chat.completion.chunk", "model": t.model,
+					"choices": []any{}, "usage": usage,
+				}))
+			}
+		}
+		out = append(out, []byte("data: [DONE]\n\n"))
+	}
+	return joinLines(out)
+}
+
+// ---------------------------------------------------------------------------
+// Chat → Responses
+
+type chatToResponses struct {
+	sseState
+	model      string
+	usage      any
+	finishSeen bool
+	done       bool
+	createdSent bool
+}
+
+func newChatToResponses() *chatToResponses { return &chatToResponses{} }
+
+func (t *chatToResponses) line(line []byte) []byte {
+	isData, data, done := t.parseLine(line)
+	if !isData {
+		return nil
+	}
+	if done {
+		// 流结束兜底：finish 后 usage 未随块到达时在此补发 completed
+		if !t.done {
+			t.done = true
+			return eventLine("response.completed", map[string]any{
+				"type": "response.completed",
+				"response": map[string]any{
+					"id": "resp_stream", "object": "response", "model": t.model, "status": "completed", "usage": t.usage,
+				},
+			})
+		}
+		return nil
+	}
+	m := eventData(data)
+	if t.model == "" {
+		t.model, _ = m["model"].(string)
+	}
+	if u, ok := m["usage"]; ok {
+		t.usage = u
+	}
+	delta := map[string]any{}
+	var finish string
+	if choices, ok := m["choices"].([]any); ok && len(choices) > 0 {
+		if c0, ok := choices[0].(map[string]any); ok {
+			if d, ok := c0["delta"].(map[string]any); ok {
+				delta = d
+			}
+			finish, _ = c0["finish_reason"].(string)
+		}
+	}
+	if finish != "" {
+		t.finishSeen = true
+	}
+	var out [][]byte
+	// 只发一次 response.created：部分上游（如 OpenRouter 的 reasoning 模型）会在
+	// 每个 chunk 的 delta 里都带 role:"assistant"，不加守卫会刷出数十条 created。
+	if !t.createdSent && delta["role"] == "assistant" {
+		t.createdSent = true
+		out = append(out, eventLine("response.created", map[string]any{
+			"type": "response.created",
+			"response": map[string]any{"id": "resp_stream", "object": "response", "model": t.model, "status": "in_progress"},
+		}))
+	}
+	if content, _ := delta["content"].(string); content != "" {
+		out = append(out, eventLine("response.output_text.delta", map[string]any{
+			"type": "response.output_text.delta", "delta": content, "item_id": "msg_1", "output_index": 0, "content_index": 0,
+		}))
+	}
+	// 上游 usage 块（choices 为空）通常晚于 finish_reason：此时再发 completed，携带 usage
+	if _, hasUsage := m["usage"]; hasUsage && t.finishSeen && !t.done {
+		t.done = true
+		out = append(out, eventLine("response.completed", map[string]any{
+			"type": "response.completed",
+			"response": map[string]any{
+				"id": "resp_stream", "object": "response", "model": t.model, "status": "completed", "usage": t.usage,
+			},
+		}))
+	}
+	return joinLines(out)
+}
diff --git a/backend/internal/proxy/convert/usage.go b/backend/internal/proxy/convert/usage.go
new file mode 100644
index 0000000..3188eaf
--- /dev/null
+++ b/backend/internal/proxy/convert/usage.go
@@ -0,0 +1,131 @@
+package convert
+
+import (
+	"encoding/json"
+)
+
+// TokenUsage 从上游响应提取的 token 用量。
+// 三种协议的字段名不同，此处统一为：input / output / cache_read / cache_creation，
+// 供用量记录与计费使用。
+type TokenUsage struct {
+	InputTokens         int
+	OutputTokens        int
+	CacheReadTokens     int
+	CacheCreationTokens int
+}
+
+// has 判断是否真的拿到了非零用量（过滤掉没有 usage 字段的响应）。
+func (u *TokenUsage) has() bool {
+	return u.InputTokens > 0 || u.OutputTokens > 0 ||
+		u.CacheReadTokens > 0 || u.CacheCreationTokens > 0
+}
+
+// mergeJSON 把一张 usage 对象并入累计值。proto 决定字段名（chat/responses 与 messages 不同）。
+func (u *TokenUsage) mergeJSON(raw map[string]any, proto string) {
+	switch proto {
+	case ProtoChat, ProtoResponses:
+		in, _ := raw["prompt_tokens"].(float64)
+		out, _ := raw["completion_tokens"].(float64)
+		if in == 0 && out == 0 {
+			in, _ = raw["input_tokens"].(float64)
+			out, _ = raw["output_tokens"].(float64)
+		}
+		u.InputTokens += int(in)
+		u.OutputTokens += int(out)
+		if d, ok := raw["prompt_tokens_details"].(map[string]any); ok {
+			if c, _ := d["cached_tokens"].(float64); c > 0 {
+				u.CacheReadTokens += int(c)
+			}
+		}
+		if d, ok := raw["input_tokens_details"].(map[string]any); ok {
+			if c, _ := d["cached_tokens"].(float64); c > 0 {
+				u.CacheReadTokens += int(c)
+			}
+		}
+	case ProtoMessages:
+		in, _ := raw["input_tokens"].(float64)
+		out, _ := raw["output_tokens"].(float64)
+		u.InputTokens += int(in)
+		u.OutputTokens += int(out)
+		if c, _ := raw["cache_read_input_tokens"].(float64); c > 0 {
+			u.CacheReadTokens += int(c)
+		}
+		if c, _ := raw["cache_creation_input_tokens"].(float64); c > 0 {
+			u.CacheCreationTokens += int(c)
+		}
+	}
+}
+
+// ExtractUsageJSON 从完整非流式响应体中提取用量。proto 为上游协议。
+// 返回 (用量, 是否有效)。
+func ExtractUsageJSON(body []byte, proto string) (TokenUsage, bool) {
+	var top map[string]any
+	if err := json.Unmarshal(body, &top); err != nil {
+		return TokenUsage{}, false
+	}
+	var u TokenUsage
+	if usage, ok := top["usage"].(map[string]any); ok {
+		u.mergeJSON(usage, proto)
+	}
+	return u, u.has()
+}
+
+// StreamUsageAccum 流式用量累计器。逐行喂入上游 SSE 的 data 载荷，
+// 按协议分别取各事件里的 usage 字段（各事件只会携带一部分字段，取最大值合并）。
+type StreamUsageAccum struct {
+	u TokenUsage
+}
+
+// NewStreamUsageAccum 创建一个流式用量累计器。
+func NewStreamUsageAccum() *StreamUsageAccum {
+	return &StreamUsageAccum{}
+}
+
+// Feed 喂入一行 SSE data 载荷（不含 "data:" 前缀与换行）。
+func (a *StreamUsageAccum) Feed(payload []byte, proto string) {
+	var top map[string]any
+	if json.Unmarshal(payload, &top) != nil {
+		return
+	}
+	var t TokenUsage
+	switch proto {
+	case ProtoChat:
+		if usage, ok := top["usage"].(map[string]any); ok {
+			t.mergeJSON(usage, proto)
+		}
+	case ProtoResponses:
+		// response.completed 事件把用量放在 response.usage 下。
+		if resp, ok := top["response"].(map[string]any); ok {
+			if usage, ok := resp["usage"].(map[string]any); ok {
+				t.mergeJSON(usage, proto)
+			}
+		}
+	case ProtoMessages:
+		// message_start: {message: {usage: {input_tokens, cache_*}}}
+		// message_delta: {usage: {output_tokens}}
+		if msg, ok := top["message"].(map[string]any); ok {
+			if usage, ok := msg["usage"].(map[string]any); ok {
+				t.mergeJSON(usage, proto)
+			}
+		}
+		if usage, ok := top["usage"].(map[string]any); ok {
+			var t2 TokenUsage
+			t2.mergeJSON(usage, proto)
+			t.InputTokens = max(t.InputTokens, t2.InputTokens)
+			t.OutputTokens = max(t.OutputTokens, t2.OutputTokens)
+			t.CacheReadTokens = max(t.CacheReadTokens, t2.CacheReadTokens)
+			t.CacheCreationTokens = max(t.CacheCreationTokens, t2.CacheCreationTokens)
+		}
+	default:
+		return
+	}
+	a.u.InputTokens = max(a.u.InputTokens, t.InputTokens)
+	a.u.OutputTokens = max(a.u.OutputTokens, t.OutputTokens)
+	a.u.CacheReadTokens = max(a.u.CacheReadTokens, t.CacheReadTokens)
+	a.u.CacheCreationTokens = max(a.u.CacheCreationTokens, t.CacheCreationTokens)
+}
+
+// Usage 返回当前累计用量。
+func (a *StreamUsageAccum) Usage() TokenUsage {
+	return a.u
+}
diff --git a/backend/internal/proxy/convert/usage_test.go b/backend/internal/proxy/convert/usage_test.go
new file mode 100644
index 0000000..90922b4
--- /dev/null
+++ b/backend/internal/proxy/convert/usage_test.go
@@ -0,0 +1,170 @@
+package convert
+
+import (
+	"encoding/json"
+	"testing"
+)
+
+// ---- ExtractUsageJSON: 非流式各协议 ----
+
+func TestExtractUsageJSONChat(t *testing.T) {
+	body := []byte(`{
+		"id": "chatcmpl-1",
+		"object": "chat.completion",
+		"choices": [{"index": 0, "message": {"role": "assistant", "content": "hi"}}],
+		"usage": {
+			"prompt_tokens": 11,
+			"completion_tokens": 7,
+			"total_tokens": 18,
+			"prompt_tokens_details": {"cached_tokens": 4}
+		}
+	}`)
+	u, ok := ExtractUsageJSON(body, ProtoChat)
+	if !ok {
+		t.Fatalf("expected ok=true")
+	}
+	if u.InputTokens != 11 || u.OutputTokens != 7 {
+		t.Fatalf("chat usage = %+v, want input=11 output=7", u)
+	}
+	if u.CacheReadTokens != 4 {
+		t.Fatalf("chat cacheRead = %d, want 4", u.CacheReadTokens)
+	}
+}
+
+func TestExtractUsageJSONMessages(t *testing.T) {
+	body := []byte(`{
+		"id": "msg_1",
+		"type": "message",
+		"role": "assistant",
+		"content": [{"type": "text", "text": "hi"}],
+		"usage": {
+			"input_tokens": 15,
+			"output_tokens": 8,
+			"cache_read_input_tokens": 3,
+			"cache_creation_input_tokens": 2
+		}
+	}`)
+	u, ok := ExtractUsageJSON(body, ProtoMessages)
+	if !ok {
+		t.Fatalf("expected ok=true")
+	}
+	if u.InputTokens != 15 || u.OutputTokens != 8 || u.CacheReadTokens != 3 || u.CacheCreationTokens != 2 {
+		t.Fatalf("messages usage = %+v", u)
+	}
+}
+
+func TestExtractUsageJSONResponses(t *testing.T) {
+	body := []byte(`{
+		"id": "resp_1",
+		"object": "response",
+		"output": [],
+		"usage": {
+			"input_tokens": 13,
+			"output_tokens": 9,
+			"input_tokens_details": {"cached_tokens": 5}
+		}
+	}`)
+	u, ok := ExtractUsageJSON(body, ProtoResponses)
+	if !ok {
+		t.Fatalf("expected ok=true")
+	}
+	if u.InputTokens != 13 || u.OutputTokens != 9 || u.CacheReadTokens != 5 {
+		t.Fatalf("responses usage = %+v", u)
+	}
+}
+
+func TestExtractUsageJSONInvalidAndMissing(t *testing.T) {
+	if _, ok := ExtractUsageJSON([]byte("not json"), ProtoChat); ok {
+		t.Fatalf("invalid json should not report ok")
+	}
+	if _, ok := ExtractUsageJSON([]byte(`{"id": "x"}`), ProtoChat); ok {
+		t.Fatalf("missing usage should not report ok")
+	}
+	// 空对象 usage：全 0 视为无效
+	if _, ok := ExtractUsageJSON([]byte(`{"usage": {}}`), ProtoChat); ok {
+		t.Fatalf("empty usage should not report ok")
+	}
+}
+
+// ---- StreamUsageAccum: 流式各协议 ----
+
+func feedLines(t *testing.T, proto string, lines ...string) TokenUsage {
+	t.Helper()
+	acc := NewStreamUsageAccum()
+	for _, ln := range lines {
+		acc.Feed([]byte(ln), proto)
+	}
+	return acc.Usage()
+}
+
+func TestStreamUsageChatFinalChunk(t *testing.T) {
+	// 前面的 chunk 不带 usage；最后一个 chunk 带完整 usage
+	u := feedLines(t, ProtoChat,
+		`{"id":"c1","object":"chat.completion.chunk","choices":[{"delta":{"content":"he"}}]}`,
+		`{"id":"c1","object":"chat.completion.chunk","choices":[{"delta":{"content":"llo"}}]}`,
+		`{"id":"c1","object":"chat.completion.chunk","choices":[],"usage":{"prompt_tokens":11,"completion_tokens":7,"prompt_tokens_details":{"cached_tokens":4}}}`,
+	)
+	if u.InputTokens != 11 || u.OutputTokens != 7 || u.CacheReadTokens != 4 {
+		t.Fatalf("chat stream usage = %+v", u)
+	}
+}
+
+func TestStreamUsageMessagesStartAndDelta(t *testing.T) {
+	// message_start 带 input/cache，message_delta 带 output；逐字段取 max 合并
+	u := feedLines(t, ProtoMessages,
+		`{"type":"message_start","message":{"id":"msg_1","usage":{"input_tokens":15,"cache_read_input_tokens":3,"cache_creation_input_tokens":2}}}`,
+		`{"type":"content_block_delta","delta":{"type":"text_delta","text":"hi"}}`,
+		`{"type":"message_delta","delta":{"stop_reason":"end_turn"},"usage":{"output_tokens":8}}`,
+	)
+	if u.InputTokens != 15 || u.OutputTokens != 8 || u.CacheReadTokens != 3 || u.CacheCreationTokens != 2 {
+		t.Fatalf("messages stream usage = %+v", u)
+	}
+}
+
+func TestStreamUsageResponsesCompleted(t *testing.T) {
+	// response.completed 事件的用量嵌在 response.usage 下
+	u := feedLines(t, ProtoResponses,
+		`{"type":"response.output_text.delta","delta":"hi"}`,
+		`{"type":"response.completed","response":{"id":"resp_1","usage":{"input_tokens":13,"output_tokens":9,"input_tokens_details":{"cached_tokens":5}}}}`,
+	)
+	if u.InputTokens != 13 || u.OutputTokens != 9 || u.CacheReadTokens != 5 {
+		t.Fatalf("responses stream usage = %+v", u)
+	}
+}
+
+func TestStreamUsageIgnoresNonDataPayloads(t *testing.T) {
+	// [DONE]、垃圾行、空对象都不应产生用量
+	u := feedLines(t, ProtoChat, `[DONE]`, `{`, ``, `{"choices":[]}`)
+	if u.has() {
+		t.Fatalf("expected zero usage, got %+v", u)
+	}
+}
+
+func TestStreamUsageFeedKeepsMaxAcrossEvents(t *testing.T) {
+	// 同一字段在多个事件出现时取较大值（防乱序/重复）
+	u := feedLines(t, ProtoMessages,
+		`{"type":"message_start","message":{"usage":{"input_tokens":15}}}`,
+		`{"type":"message_delta","usage":{"output_tokens":5}}`,
+		`{"type":"message_delta","usage":{"output_tokens":8}}`,
+	)
+	if u.InputTokens != 15 || u.OutputTokens != 8 {
+		t.Fatalf("max-merge usage = %+v", u)
+	}
+}
+
+// ---- usage JSON 结构合法性（防止手写 struct 漂移）----
+
+func TestUsageJSONRoundTrip(t *testing.T) {
+	u := TokenUsage{InputTokens: 10, OutputTokens: 5, CacheReadTokens: 2, CacheCreationTokens: 1}
+	b, err := json.Marshal(u)
+	if err != nil {
+		t.Fatalf("marshal: %v", err)
+	}
+	var back TokenUsage
+	if err := json.Unmarshal(b, &back); err != nil {
+		t.Fatalf("unmarshal: %v", err)
+	}
+	if back != u {
+		t.Fatalf("round trip = %+v, want %+v", back, u)
+	}
+}
diff --git a/backend/internal/proxy/gateway.go b/backend/internal/proxy/gateway.go
index 1689d48..0b74c72 100644
--- a/backend/internal/proxy/gateway.go
+++ b/backend/internal/proxy/gateway.go
@@ -1,6 +1,7 @@
 package proxy
 
 import (
+	"bufio"
 	"bytes"
 	"context"
 	"encoding/json"
@@ -13,6 +14,7 @@ import (
 	"opencatd-open/internal/dao"
 	"opencatd-open/internal/proxy/convert"
 	"opencatd-open/internal/store"
+	"opencatd-open/internal/usage"
 	"opencatd-open/pkg/config"
 	"os"
 	"strings"
@@ -35,6 +37,7 @@ type Gateway struct {
 	usageDAO   *dao.UsageDAO
 	dailyDAO   *dao.DailyUsageDAO
 	channelSvc *channel.Service
+	usageRec   *usage.Recorder
 }
 
 func NewGateway(ctx context.Context, cfg *config.Config, db *gorm.DB, wg *sync.WaitGroup, userDAO *dao.UserDAO, apiKeyDAO *dao.ApiKeyDAO, usageDAO *dao.UsageDAO, dailyDAO *dao.DailyUsageDAO) *Gateway {
@@ -67,6 +70,11 @@ func (g *Gateway) SetChannelService(svc *channel.Service) {
 	g.channelSvc = svc
 }
 
+// SetUsageRecorder 注入异步用量记录器；nil 时网关跳过用量上报。
+func (g *Gateway) SetUsageRecorder(r *usage.Recorder) {
+	g.usageRec = r
+}
+
 // Request represents a parsed incoming request
 type Request struct {
 	Model     string
@@ -144,26 +152,24 @@ func (g *Gateway) Dispatch(c *gin.Context, req *Request) {
 		return
 	}
 
-	// Determine target format and convert if needed
-	targetFormat := req.Protocol
-	if len(ch.FormatsEffective()) > 0 {
-		// Prefer the channel's native format
-		for _, f := range ch.FormatsEffective() {
-			if f == req.Protocol {
-				targetFormat = f
-				break
-			}
-		}
+	// Determine target format: channel declares support for the client protocol
+	// then passthrough, otherwise convert to its first supported protocol
+	// (chat > messages > responses).
+	targetFormat := g.conversionTarget(ch, req.Protocol)
+	if targetFormat == "" {
+		g.writeError(c, http.StatusBadGateway, fmt.Sprintf("channel %q declares no supported protocol format", ch.Name))
+		return
 	}
 
 	// Build upstream URL
-	upstreamPath := g.getUpstreamPath(req.Protocol)
-	upstreamURL := ch.UpstreamURL(req.Protocol, upstreamPath)
+	upstreamPath := g.getUpstreamPath(targetFormat)
+	upstreamURL := ch.UpstreamURL(targetFormat, upstreamPath)
 
 	// Convert request if needed
 	var requestBody []byte
 	if targetFormat != req.Protocol {
-		requestBody, err = g.convertRequest(req.Body, req.Protocol, targetFormat)
+		var err error
+		requestBody, err = convert.ConvertRequest(req.Body, req.Protocol, targetFormat)
 		if err != nil {
 			g.writeError(c, http.StatusBadRequest, "conversion failed: "+err.Error())
 			return
@@ -206,12 +212,31 @@ func (g *Gateway) Dispatch(c *gin.Context, req *Request) {
 
 	// Stream or buffer response
 	if req.Stream {
-		g.streamResponse(c, resp, req.Protocol, ch)
+		g.streamResponse(c, resp, req.Protocol, targetFormat)
 	} else {
-		g.bufferResponse(c, resp, req.Protocol, ch)
+		g.bufferResponse(c, resp, req.Protocol, targetFormat)
 	}
 }
 
+// conversionTarget 决定客户端协议在渠道上的处理方式：
+// 渠道声明支持该协议则直通；否则转为其首选支持协议（chat > messages > responses）。
+func (g *Gateway) conversionTarget(ch *store.Channel, clientProto string) string {
+	formats := ch.FormatsEffective()
+	for _, f := range formats {
+		if f == clientProto {
+			return clientProto
+		}
+	}
+	for _, p := range []string{convert.ProtoChat, convert.ProtoMessages, convert.ProtoResponses} {
+		for _, f := range formats {
+			if f == p {
+				return p
+			}
+		}
+	}
+	return ""
+}
+
 func (g *Gateway) getUpstreamPath(protocol string) string {
 	switch protocol {
 	case "chat":
@@ -237,116 +262,86 @@ func (g *Gateway) setHeaders(req *http.Request, ch *store.Channel, apiKey string
 	}
 }
 
-func (g *Gateway) convertRequest(body []byte, from, to string) ([]byte, error) {
-	switch {
-	case from == "chat" && to == "messages":
-		var req convert.ChatCompletionRequest
-		if err := json.Unmarshal(body, &req); err != nil {
-			return nil, err
-		}
-		msgReq, err := convert.ChatToMessages(&req)
-		if err != nil {
-			return nil, err
-		}
-		return json.Marshal(msgReq)
 
-	case from == "chat" && to == "responses":
-		var req convert.ChatCompletionRequest
-		if err := json.Unmarshal(body, &req); err != nil {
-			return nil, err
-		}
-		respReq, err := convert.ChatToResponses(&req)
-		if err != nil {
-			return nil, err
-		}
-		return json.Marshal(respReq)
-
-	case from == "messages" && to == "chat":
-		var req convert.MessagesRequest
-		if err := json.Unmarshal(body, &req); err != nil {
-			return nil, err
-		}
-		// Messages -> Chat: we need to construct a ChatCompletionRequest
-		chatReq := &convert.ChatCompletionRequest{
-			Model: req.Model,
-		}
-		for _, m := range req.Messages {
-			chatReq.Messages = append(chatReq.Messages, m)
-		}
-		if req.Temperature != nil {
-			chatReq.Temperature = req.Temperature
-		}
-		if req.TopP != nil {
-			chatReq.TopP = req.TopP
-		}
-		chatReq.Tools = req.Tools
-		chatReq.Stream = req.Stream
-		return json.Marshal(chatReq)
-
-	case from == "responses" && to == "chat":
-		var req convert.ResponsesRequest
-		if err := json.Unmarshal(body, &req); err != nil {
-			return nil, err
-		}
-		chatReq := &convert.ChatCompletionRequest{
-			Model: req.Model,
-		}
-		for _, item := range req.Input {
-			chatReq.Messages = append(chatReq.Messages, convert.Message{
-				Role:    item.Role,
-				Content: item.Content,
-			})
-		}
-		chatReq.Tools = req.Tools
-		chatReq.Stream = req.Stream
-		return json.Marshal(chatReq)
-
-	default:
-		return body, nil
-	}
-}
-
-func (g *Gateway) streamResponse(c *gin.Context, resp *http.Response, protocol string, ch *store.Channel) {
+// streamResponse 流式响应：按 \n\n 分块零缓冲转发；跨协议时逐行转换。
+func (g *Gateway) streamResponse(c *gin.Context, resp *http.Response, clientProto, upstreamProto string) {
+	w := c.Writer
 	c.Header("Content-Type", "text/event-stream")
 	c.Header("Cache-Control", "no-cache")
 	c.Header("Connection", "keep-alive")
 	c.Status(http.StatusOK)
 
-	writer := convert.NewSSEWriter(c.Writer)
-	parser := convert.NewSSEParser(resp.Body)
+	flusher, _ := w.(http.Flusher)
 
-	for {
-		event, err := parser.ReadEvent()
-		if err != nil {
-			if err == io.EOF {
-				break
-			}
-			log.Printf("Stream parse error: %v", err)
-			break
-		}
-
-		if event.Event == "error" {
-			log.Printf("Upstream stream error: %s", event.Data)
-			break
-		}
-
-		// Write raw SSE event based on protocol
-		if err := writer.WriteEvent("chat CompletionChunk", event.Data); err != nil {
-			break
-		}
+	// 跨协议时按行转换；同协议直通（lineConv 为 nil）。
+	var lineConv func([]byte) []byte
+	if upstreamProto != clientProto {
+		lineConv = convert.NewStreamTransformer(upstreamProto, clientProto)
 	}
 
-	writer.WriteDone()
+	// 上游原始行按 \n\n 分块，避免把 data 行内的转义换行当成事件边界。
+	r := bufio.NewReaderSize(resp.Body, 32*1024)
+	for {
+		buf := []byte{}
+		for {
+			line, err := r.ReadSlice('\n')
+			if err == bufio.ErrBufferFull {
+				buf = append(buf, line...)
+				continue
+			}
+			buf = append(buf, line...)
+			if err == io.EOF {
+				if len(buf) == 0 {
+					return
+				}
+				if !bytes.HasSuffix(buf, []byte("\n")) {
+					buf = append(buf, '\n')
+				}
+			} else if err != nil {
+				log.Printf("stream read error: %v", err)
+				return
+			}
+			if len(buf) >= 2 && bytes.HasSuffix(buf, []byte("\n\n")) {
+				break
+			}
+		}
+
+		out := buf
+		if lineConv != nil {
+			out = lineConv(buf)
+		}
+		if len(out) == 0 {
+			continue
+		}
+		if _, err := w.Write(out); err != nil {
+			return // 客户端已断开
+		}
+		if flusher != nil {
+			flusher.Flush()
+		}
+	}
 }
 
-func (g *Gateway) bufferResponse(c *gin.Context, resp *http.Response, protocol string, ch *store.Channel) {
+func (g *Gateway) bufferResponse(c *gin.Context, resp *http.Response, clientProto, upstreamProto string) {
 	body, err := io.ReadAll(resp.Body)
 	if err != nil {
 		g.writeError(c, http.StatusBadGateway, "failed to read response")
 		return
 	}
 
-	c.Data(resp.StatusCode, "application/json", body)
+	out := body
+	if upstreamProto != clientProto {
+		if converted, cerr := convert.ConvertResponse(body, upstreamProto, clientProto); cerr == nil {
+			out = converted
+		} else {
+			// 转换失败时至少剥掉非 JSON 前缀，让客户端能解析出正文
+			out = convert.CleanJSON(body)
+		}
+	} else {
+		// 直通：部分上游（如 OpenRouter）的 non-stream 响应在 JSON 前夹带空白/注释
+		out = convert.CleanJSON(body)
+	}
+	c.Data(resp.StatusCode, "application/json", out)
 }
 
 func (g *Gateway) writeError(c *gin.Context, status int, message string) {
diff --git a/backend/internal/usage/recorder.go b/backend/internal/usage/recorder.go
index 5b45e6a..074512e 100644
--- a/backend/internal/usage/recorder.go
+++ b/backend/internal/usage/recorder.go
@@ -2,6 +2,7 @@ package usage
 
 import (
 	"context"
+	"fmt"
 	"log"
 	"opencatd-open/internal/dao"
 	"opencatd-open/internal/store"
@@ -11,16 +12,26 @@ import (
 
 // Event represents a usage event to be recorded
 type Event struct {
-	UserID           uint64
-	ModelName        string
-	ChannelID        uint64
-	PromptTokens     int
-	CompletionTokens int
-	CacheReadTokens  int
-	Cost             float64
-	IsError          bool
-	IsCanceled       bool
-	RequestID        string
+	UserID              uint64
+	ModelName           string
+	ChannelID           uint64
+	PromptTokens        int
+	CompletionTokens    int
+	CacheReadTokens     int
+	CacheCreationTokens int
+	Cost                float64
+	IsError             bool
+	IsCanceled          bool
+	RequestID           string
+	KeyID               uint64
+	Protocol            string
+	ErrorCode           string
+	LatencyMS           int
+	InputPrice          float64
+	OutputPrice         float64
+	CacheReadPrice      float64
+	TraceID             string // TraceID for distributed tracing
+	ModelID             uint64 // Model ID from channel-model binding
 }
 
 // Recorder handles async usage recording
@@ -117,16 +128,31 @@ func (r *Recorder) flush(events []Event) {
 			status = store.UsageStatusCanceled
 		}
 
+		var errCode *string
+		if e.ErrorCode != "" {
+			errCode = &e.ErrorCode
+		}
+
 		log := &store.UsageLog{
-			UserID:           e.UserID,
-			ModelName:        e.ModelName,
-			ChannelID:        e.ChannelID,
-			InputTokens:      int64(e.PromptTokens),
-			OutputTokens:     int64(e.CompletionTokens),
-			CacheReadTokens:  int64(e.CacheReadTokens),
-			Cost:             e.Cost,
-			Status:           status,
-			RequestID:        e.RequestID,
+			UserID:              e.UserID,
+			KeyID:               e.KeyID,
+			ChannelID:           e.ChannelID,
+			ModelID:             e.ModelID,
+			ModelName:           e.ModelName,
+			Protocol:            e.Protocol,
+			InputTokens:         int64(e.PromptTokens),
+			OutputTokens:        int64(e.CompletionTokens),
+			CacheReadTokens:     int64(e.CacheReadTokens),
+			CacheCreationTokens: int64(e.CacheCreationTokens),
+			InputPrice:          e.InputPrice,
+			OutputPrice:         e.OutputPrice,
+			CacheReadPrice:      e.CacheReadPrice,
+			Cost:                e.Cost,
+			LatencyMS:           e.LatencyMS,
+			Status:              status,
+			ErrorCode:           errCode,
+			RequestID:           e.RequestID,
+			TraceID:             e.TraceID,
 		}
 		logs = append(logs, log)
 	}
@@ -136,5 +162,39 @@ func (r *Recorder) flush(events []Event) {
 		log.Printf("Failed to batch create usage logs: %v", err)
 	}
 
+	// Daily rollup for success and canceled requests
+	dailyMap := make(map[string]*store.UsageDaily)
+	for _, e := range events {
+		if e.IsError {
+			continue
+		}
+		date := time.Now().Format("2006-01-02")
+		key := fmt.Sprintf("%d:%d:%s", e.UserID, e.ModelID, date)
+		d := dailyMap[key]
+		if d == nil {
+			d = &store.UsageDaily{
+				UserID:          e.UserID,
+				ModelID:         e.ModelID,
+				Date:            date,
+				Requests:        0,
+				InputTokens:     0,
+				OutputTokens:    0,
+				CacheReadTokens: 0,
+				Cost:            0,
+			}
+			dailyMap[key] = d
+		}
+		d.Requests++
+		d.InputTokens += int64(e.PromptTokens)
+		d.OutputTokens += int64(e.CompletionTokens)
+		d.CacheReadTokens += int64(e.CacheReadTokens)
+		d.Cost += e.Cost
+	}
+	for _, d := range dailyMap {
+		if err := r.dailyDAO.UpsertDailyUsage(context.Background(), d); err != nil {
+			log.Printf("Failed to upsert daily usage: %v", err)
+		}
+	}
+
 	log.Printf("Flushed %d usage logs", len(logs))
 }
diff --git a/backend/middleware/auth_llm.go b/backend/middleware/auth_llm.go
index b21e618..f33cc51 100644
--- a/backend/middleware/auth_llm.go
+++ b/backend/middleware/auth_llm.go
@@ -9,45 +9,34 @@ import (
 	"gorm.io/gorm"
 )
 
+// keyPrefixLen 是 key_prefix 列的截断长度，必须与 api.go:459 的 keyValue[:12] 一致。
+// 真实 key 为 sk-ot- + 48 位 hex（54 字符），故 12 位足够唯一。
+const keyPrefixLen = 12
+
 func AuthLLM(db *gorm.DB) gin.HandlerFunc {
 	return func(c *gin.Context) {
-		authToken := c.GetHeader("Authorization")
-		if authToken == "" {
-			c.AbortWithStatusJSON(http.StatusUnauthorized, gin.H{
-				"error": map[string]interface{}{
-					"message": "未提供认证信息",
-					"type":    "invalid_request_error",
-				},
-			})
+		key := extractAPIKey(c.GetHeader("Authorization"))
+
+		// 区分「没传」和「传了但不对」，便于排查客户端配置。
+		if strings.TrimSpace(c.GetHeader("Authorization")) == "" {
+			unauthorized(c, "未提供认证信息")
+			return
+		}
+		// 长度不足时直接拒绝：避免下方 authToken[:12] 越界 panic 打崩进程。
+		if len(key) < keyPrefixLen {
+			unauthorized(c, "无效的API密钥")
 			return
 		}
 
-		// Extract API key from Bearer token
-		if len(authToken) > 7 {
-			authToken = authToken[7:]
-		}
-
-		// Find API key by prefix
 		var apiKey store.APIKey
-		if err := db.Where("key_prefix = ? AND status = ?", authToken[:8], store.KeyStatusActive).First(&apiKey).Error; err != nil {
-			c.AbortWithStatusJSON(http.StatusUnauthorized, gin.H{
-				"error": map[string]interface{}{
-					"message": "无效的API密钥",
-					"type":    "invalid_request_error",
-				},
-			})
+		if err := db.Where("key_prefix = ? AND status = ?", key[:keyPrefixLen], store.KeyStatusActive).First(&apiKey).Error; err != nil {
+			unauthorized(c, "无效的API密钥")
 			return
 		}
 
 		// Verify full key hash
-		keyHash := store.HashAPIKey(authToken)
-		if apiKey.KeyHash != keyHash {
-			c.AbortWithStatusJSON(http.StatusUnauthorized, gin.H{
-				"error": map[string]interface{}{
-					"message": "无效的API密钥",
-					"type":    "invalid_request_error",
-				},
-			})
+		if apiKey.KeyHash != store.HashAPIKey(key) {
+			unauthorized(c, "无效的API密钥")
 			return
 		}
 
@@ -57,11 +46,20 @@ func AuthLLM(db *gorm.DB) gin.HandlerFunc {
 	}
 }
 
-// extractAPIKey extracts the API key from the Authorization header
-func extractAPIKey(c *gin.Context) string {
-	auth := c.GetHeader("Authorization")
+// extractAPIKey 从 Authorization 头取 Bearer token，兼容无 "Bearer " 前缀的直传。
+func extractAPIKey(auth string) string {
+	auth = strings.TrimSpace(auth)
 	if strings.HasPrefix(auth, "Bearer ") {
-		return auth[7:]
+		return strings.TrimSpace(auth[len("Bearer "):])
 	}
 	return auth
 }
+
+func unauthorized(c *gin.Context, message string) {
+	c.AbortWithStatusJSON(http.StatusUnauthorized, gin.H{
+		"error": map[string]interface{}{
+			"message": message,
+			"type":    "invalid_request_error",
+		},
+	})
+}
diff --git a/backend/router/setRouter.go b/backend/router/setRouter.go
index c43eb3e..8b71fe1 100644
--- a/backend/router/setRouter.go
+++ b/backend/router/setRouter.go
@@ -50,7 +50,7 @@ func SetRouter(cfg *config.Config, db *gorm.DB, web *embed.FS) {
 
 	// Initialize health checker and start periodic checks
 	healthChecker := channel.NewHealthChecker(channelDAO, channelSvc)
-	go healthChecker.StartPeriodicCheck(ctx, 5*time.Minute)
+	go healthChecker.StartPeriodicCheck(ctx)
 
 	// Initialize usage recorder and start background worker
 	usageRecorder := usage.NewRecorder(usageDAO, dailyDAO)
@@ -60,7 +60,7 @@ func SetRouter(cfg *config.Config, db *gorm.DB, web *embed.FS) {
 	// Initialize gateway
 	gateway := proxy.NewGateway(ctx, cfg, db, &wg, userDAO, apiKeyDAO, usageDAO, dailyDAO)
 	gateway.SetChannelService(channelSvc)
-
+	gateway.SetUsageRecorder(usageRecorder)
 	// Initialize API handler
 	apiHandler := api.NewHandler(db)
 
@@ -99,7 +99,7 @@ func SetRouter(cfg *config.Config, db *gorm.DB, web *embed.FS) {
 		apiGroup.DELETE("/keys/:id", apiHandler.DeleteApiKey)
 		apiGroup.POST("/keys/batch/:option", apiHandler.BatchApiKeys)
 
-		// Channel management
+		// Channel management (legacy endpoints)
 		apiGroup.GET("/channels", apiHandler.ListChannels)
 		apiGroup.POST("/channels", apiHandler.CreateChannel)
 		apiGroup.PUT("/channels/:id", apiHandler.UpdateChannel)
@@ -107,11 +107,40 @@ func SetRouter(cfg *config.Config, db *gorm.DB, web *embed.FS) {
 		apiGroup.GET("/channels/:id/models", apiHandler.GetChannelModels)
 		apiGroup.POST("/channels/:id/models", apiHandler.BindChannelModels)
 
-		// Model management
+		// Model management (legacy endpoints)
 		apiGroup.GET("/models", apiHandler.ListModels)
 		apiGroup.POST("/models", apiHandler.CreateModel)
 		apiGroup.PUT("/models/:id", apiHandler.UpdateModel)
 		apiGroup.DELETE("/models/:id", apiHandler.DeleteModel)
+
+		// Admin channel management (enhanced)
+		apiGroup.GET("/admin/channels", apiHandler.AdminChannels)
+		apiGroup.POST("/admin/channels", apiHandler.AdminCreateChannel)
+		apiGroup.PUT("/admin/channels/:id", apiHandler.AdminUpdateChannel)
+		apiGroup.DELETE("/admin/channels/:id", apiHandler.AdminDeleteChannel)
+		apiGroup.POST("/admin/channels/:id/test", apiHandler.AdminTestChannel)
+		apiGroup.GET("/admin/channels/:id/models/remote", apiHandler.AdminChannelRemoteModels)
+		apiGroup.GET("/admin/channels/:id/models", apiHandler.AdminChannelModels)
+		apiGroup.POST("/admin/channels/:id/models", apiHandler.AdminChannelAddModel)
+		apiGroup.PATCH("/admin/channels/:id/models/:bid", apiHandler.AdminChannelUpdateModel)
+		apiGroup.DELETE("/admin/channels/:id/models/:bid", apiHandler.AdminChannelDeleteModel)
+
+		// Admin model management (enhanced)
+		apiGroup.GET("/admin/models", apiHandler.AdminModels)
+		apiGroup.DELETE("/admin/models/unused", apiHandler.AdminDeleteUnusedModels)
+		apiGroup.POST("/admin/models", apiHandler.AdminCreateModel)
+		apiGroup.PUT("/admin/models/:id", apiHandler.AdminUpdateModel)
+		apiGroup.DELETE("/admin/models/:id", apiHandler.AdminDeleteModel)
+		apiGroup.POST("/admin/models/:id/bindings", apiHandler.AdminCreateModelBinding)
+		apiGroup.DELETE("/admin/models/:id/bindings/:bid", apiHandler.AdminDeleteModelBinding)
+
+		// Admin system config
+		apiGroup.GET("/admin/config", apiHandler.AdminGetConfig)
+		apiGroup.PUT("/admin/config", apiHandler.AdminUpdateConfig)
+		apiGroup.GET("/admin/config/registration", apiHandler.AdminGetRegistration)
+		apiGroup.PUT("/admin/config/registration", apiHandler.AdminUpdateRegistration)
+		apiGroup.GET("/admin/config/password-login", apiHandler.AdminGetPasswordLogin)
+		apiGroup.PUT("/admin/config/password-login", apiHandler.AdminUpdatePasswordLogin)
 	}
 
 	// LLM proxy routes
diff --git a/frontend/src/components/common/Toast.vue b/frontend/src/components/common/Toast.vue
index acb1dde..cfcf79d 100644
--- a/frontend/src/components/common/Toast.vue
+++ b/frontend/src/components/common/Toast.vue
@@ -47,9 +47,9 @@ const iconForType = (type: ToastType) => {
 const typeClasses = (type: ToastType) => {
   switch (type) {
     case 'success':
-      return 'border-success/20 bg-success/10 text-success-content dark:border-success/30 dark:bg-success/15';
+      return 'border-success bg-success/15 text-success';
     case 'error':
-      return 'border-error/20 bg-error/10 text-error-content dark:border-error/30 dark:bg-error/15';
+      return 'border-error bg-error/15 text-error';
     default:
       return 'border-base-300 bg-base-100 text-base-content';
   }
diff --git a/frontend/src/components/ui/Badge.vue b/frontend/src/components/ui/Badge.vue
new file mode 100644
index 0000000..209ae69
--- /dev/null
+++ b/frontend/src/components/ui/Badge.vue
@@ -0,0 +1,30 @@
+<script setup lang="ts">
+withDefaults(defineProps<{ variant?: 'neutral' | 'ok' | 'warn' | 'err' | 'accent' }>(), {
+  variant: 'neutral',
+})
+</script>
+
+<template>
+  <span
+    class="inline-flex items-center gap-1.5 rounded-full px-2 py-0.5 text-[11px] leading-5"
+    :class="{
+      neutral: 'bg-base-200 text-base-content',
+      ok: 'bg-success/10 text-success',
+      warn: 'bg-warning/10 text-warning',
+      err: 'bg-error/10 text-error',
+      accent: 'bg-primary/10 text-primary',
+    }[variant]"
+  >
+    <span
+      v-if="variant !== 'neutral'"
+      class="size-1.5 rounded-full"
+      :class="{
+        ok: 'bg-success',
+        warn: 'bg-warning',
+        err: 'bg-error',
+        accent: 'bg-primary',
+      }[variant]"
+    />
+    <slot />
+  </span>
+</template>
diff --git a/frontend/src/components/ui/Button.vue b/frontend/src/components/ui/Button.vue
new file mode 100644
index 0000000..ed5d7b1
--- /dev/null
+++ b/frontend/src/components/ui/Button.vue
@@ -0,0 +1,27 @@
+<script setup lang="ts">
+withDefaults(
+  defineProps<{
+    variant?: 'primary' | 'ghost' | 'danger'
+    size?: 'sm' | 'md'
+    loading?: boolean
+    disabled?: boolean
+  }>(),
+  { variant: 'primary', size: 'md', loading: false, disabled: false },
+)
+</script>
+
+<template>
+  <button
+    :disabled="disabled || loading"
+    class="inline-flex items-center justify-center gap-2 rounded-md font-medium transition-[transform,background-color,border-color,color] duration-150 active:scale-[0.98] disabled:pointer-events-none disabled:opacity-50 select-none"
+    :class="[
+      size === 'sm' ? 'h-8 px-3 text-xs' : 'h-10 px-4 text-sm',
+      variant === 'primary' && 'bg-primary text-primary-content hover:bg-primary/90',
+      variant === 'ghost' && 'border border-base-300/60 text-base-content hover:bg-base-200/50',
+      variant === 'danger' && 'border border-error text-error hover:bg-error/10',
+    ]"
+  >
+    <span v-if="loading" class="size-3.5 animate-spin rounded-full border-2 border-current border-t-transparent" />
+    <slot />
+  </button>
+</template>
diff --git a/frontend/src/components/ui/Input.vue b/frontend/src/components/ui/Input.vue
new file mode 100644
index 0000000..b519f3f
--- /dev/null
+++ b/frontend/src/components/ui/Input.vue
@@ -0,0 +1,36 @@
+<script setup lang="ts">
+withDefaults(
+  defineProps<{
+    label?: string
+    modelValue?: string | number
+    type?: string
+    placeholder?: string
+    hint?: string
+    error?: string
+    autocomplete?: string
+    disabled?: boolean
+    maxlength?: number
+  }>(),
+  { type: 'text', modelValue: '', disabled: false },
+)
+const emit = defineEmits<{ 'update:modelValue': [string | number] }>()
+</script>
+
+<template>
+  <label class="block">
+    <span v-if="label" class="mb-1.5 block text-xs font-medium text-base-content/50">{{ label }}</span>
+    <input
+      :type="type"
+      :value="modelValue"
+      :placeholder="placeholder"
+      :autocomplete="autocomplete"
+      :disabled="disabled"
+      :maxlength="maxlength"
+      class="h-10 w-full rounded-md border border-base-300/60 bg-base-100 px-3 text-sm text-base-content placeholder-base-content/40 outline-none transition focus:border-primary focus:ring-2 focus:ring-primary disabled:cursor-not-allowed disabled:opacity-50"
+      :class="error && 'border-error focus:border-error focus:ring-error'"
+      @input="emit('update:modelValue', ($event.target as HTMLInputElement).value as string | number)"
+    />
+    <span v-if="hint && !error" class="mt-1.5 block text-xs text-base-content/50">{{ hint }}</span>
+    <span v-if="error" class="mt-1.5 block text-xs text-error">{{ error }}</span>
+  </label>
+</template>
diff --git a/frontend/src/components/ui/Modal.vue b/frontend/src/components/ui/Modal.vue
new file mode 100644
index 0000000..4306095
--- /dev/null
+++ b/frontend/src/components/ui/Modal.vue
@@ -0,0 +1,84 @@
+<script setup lang="ts">
+import { nextTick, onMounted, onUnmounted, ref, watch } from 'vue'
+import { X } from '@lucide/vue'
+
+const props = withDefaults(
+  defineProps<{
+    open: boolean
+    title?: string
+    width?: string
+  }>(),
+  { width: 'max-w-md' },
+)
+const emit = defineEmits<{ close: [] }>()
+
+const panel = ref<HTMLElement | null>(null)
+
+function onKey(e: KeyboardEvent) {
+  if (e.key === 'Escape' && props.open) emit('close')
+}
+onMounted(() => window.addEventListener('keydown', onKey))
+onUnmounted(() => window.removeEventListener('keydown', onKey))
+
+watch(
+  () => props.open,
+  async (open) => {
+    document.body.style.overflow = open ? 'hidden' : ''
+    if (open) {
+      await nextTick()
+      panel.value?.focus()
+    }
+  },
+)
+onUnmounted(() => {
+  document.body.style.overflow = ''
+})
+</script>
+
+<template>
+  <Teleport to="body">
+    <Transition
+      enter-active-class="transition-opacity duration-150"
+      enter-from-class="opacity-0"
+      leave-active-class="transition-opacity duration-150"
+      leave-to-class="opacity-0"
+    >
+      <div
+        v-if="open"
+        class="fixed inset-0 z-50 flex items-start justify-center overflow-y-auto bg-black/60 p-4 pt-[12vh] backdrop-blur-sm"
+        @mousedown.self="emit('close')"
+      >
+        <Transition
+          enter-active-class="transition-transform duration-150"
+          enter-from-class="scale-[0.97] opacity-0"
+          leave-active-class="transition-transform duration-150"
+          leave-to-class="scale-[0.97] opacity-0"
+        >
+          <div
+            v-if="open"
+            ref="panel"
+            role="dialog"
+            aria-modal="true"
+            :aria-label="title || '对话框'"
+            tabindex="-1"
+            class="card w-full bg-base-100 shadow-xl outline-none"
+            :class="width"
+          >
+            <div class="flex items-center justify-between border-b border-base-300/60 px-5 py-3.5">
+              <h3 class="text-sm font-semibold text-base-content">{{ title }}</h3>
+              <button class="rounded-md p-1 text-base-content/40 hover:bg-base-200 hover:text-base-content" aria-label="关闭" @click="emit('close')">
+                <X :size="16" />
+              </button>
+            </div>
+            <div class="px-5 py-4">
+              <slot />
+            </div>
+            <div v-if="$slots.footer" class="flex justify-end gap-2 border-t border-base-300/60 px-5 py-3.5">
+              <slot name="footer" />
+            </div>
+          </div>
+        </Transition>
+      </div>
+    </Transition>
+  </Teleport>
+</template>
diff --git a/frontend/src/lib/protocol.ts b/frontend/src/lib/protocol.ts
new file mode 100644
index 0000000..40cb03c
--- /dev/null
+++ b/frontend/src/lib/protocol.ts
@@ -0,0 +1,27 @@
+// 协议格式显示名与选项
+export const PROTOCOL_NAMES: Record<string, string> = {
+  chat: 'OpenAI Chat Completions',
+  responses: 'OpenAI Responses API',
+  messages: 'Anthropic Messages',
+}
+
+// 渠道表格用的短标识
+export const PROTOCOL_SHORT: Record<string, string> = {
+  chat: 'chat/completions',
+  responses: 'responses',
+  messages: 'messages',
+}
+
+export function protocolShort(p: string): string {
+  return PROTOCOL_SHORT[p] ?? p
+}
+
+export const PROTOCOL_OPTIONS: { value: string; label: string }[] = [
+  { value: 'chat', label: 'OpenAI Chat Completions' },
+  { value: 'responses', label: 'OpenAI Responses API' },
+  { value: 'messages', label: 'Anthropic Messages' },
+]
+
+export function protocolName(p: string): string {
+  return PROTOCOL_NAMES[p] ?? p
+}
diff --git a/frontend/src/stores/channel.ts b/frontend/src/stores/channel.ts
index febebfc..9e2f5d9 100644
--- a/frontend/src/stores/channel.ts
+++ b/frontend/src/stores/channel.ts
@@ -8,6 +8,8 @@ export type Channel = {
   name: string
   provider: string
   base_url: string
+  base_urls?: Record<string, string>
+  api_key_masked?: string
   weight: number
   priority: number
   timeout_ms: number
@@ -31,6 +33,14 @@ export type NewChannelPayload = {
   formats?: string[]
 }
 
+export type ChannelModelBinding = {
+  id: number
+  model_id: number
+  model_name: string
+  upstream_model: string
+  weight: number
+}
+
 export const useChannelStore = defineStore('channel', () => {
   const loading = ref(false);
   const error = ref<string | null>(null);
@@ -125,6 +135,91 @@ export const useChannelStore = defineStore('channel', () => {
     }
   };
 
+  // Admin API methods
+  const testChannel = async (id: number | string) => {
+    loading.value = true;
+    error.value = null;
+    try {
+      const response: AxiosResponse = await request.post(`/admin/channels/${id}/test`);
+      return response;
+    } catch (err: any) {
+      error.value = err.response?.data?.error || 'Failed to test channel';
+      throw err;
+    } finally {
+      loading.value = false;
+    }
+  };
+
+  const fetchRemoteModels = async (id: number | string) => {
+    loading.value = true;
+    error.value = null;
+    try {
+      const response = await request.get(`/admin/channels/${id}/models/remote`);
+      return response.data.data ?? [];
+    } catch (err: any) {
+      error.value = err.response?.data?.error || 'Failed to fetch remote models';
+      throw err;
+    } finally {
+      loading.value = false;
+    }
+  };
+
+  const fetchChannelModels = async (id: number | string) => {
+    loading.value = true;
+    error.value = null;
+    try {
+      const response = await request.get(`/admin/channels/${id}/models`);
+      return response.data.data ?? [];
+    } catch (err: any) {
+      error.value = err.response?.data?.error || 'Failed to fetch channel models';
+      throw err;
+    } finally {
+      loading.value = false;
+    }
+  };
+
+  const addChannelModel = async (id: number | string, data: { model_id: number; upstream_model: string; weight?: number }) => {
+    loading.value = true;
+    error.value = null;
+    try {
+      const response: AxiosResponse = await request.post(`/admin/channels/${id}/models`, data);
+      return response;
+    } catch (err: any) {
+      error.value = err.response?.data?.error || 'Failed to add model';
+      throw err;
+    } finally {
+      loading.value = false;
+    }
+  };
+
+  const updateChannelModel = async (channelId: number | string, bindingId: number | string, data: { upstream_model?: string; weight?: number }) => {
+    loading.value = true;
+    error.value = null;
+    try {
+      const response: AxiosResponse = await request.patch(`/admin/channels/${channelId}/models/${bindingId}`, data);
+      return response;
+    } catch (err: any) {
+      error.value = err.response?.data?.error || 'Failed to update binding';
+      throw err;
+    } finally {
+      loading.value = false;
+    }
+  };
+
+  const deleteChannelModel = async (channelId: number | string, bindingId: number | string) => {
+    loading.value = true;
+    error.value = null;
+    try {
+      const response: AxiosResponse = await request.delete(`/admin/channels/${channelId}/models/${bindingId}`);
+      return response;
+    } catch (err: any) {
+      error.value = err.response?.data?.error || 'Failed to delete binding';
+      throw err;
+    } finally {
+      loading.value = false;
+    }
+  };
+
   return {
     loading, error,
     channel, channels, totalChannels,
@@ -134,5 +229,11 @@ export const useChannelStore = defineStore('channel', () => {
     updateChannel,
     deleteChannel,
     batchChannels,
+    testChannel,
+    fetchRemoteModels,
+    fetchChannelModels,
+    addChannelModel,
+    updateChannelModel,
+    deleteChannelModel,
   };
 });
diff --git a/frontend/src/stores/model.ts b/frontend/src/stores/model.ts
new file mode 100644
index 0000000..c09adde
--- /dev/null
+++ b/frontend/src/stores/model.ts
@@ -0,0 +1,171 @@
+import { defineStore } from 'pinia';
+import { ref } from 'vue';
+import type { AxiosResponse } from 'axios';
+import request from '@/api/client';
+
+export type Model = {
+  id: number
+  name: string
+  display_name?: string
+  input_price: number
+  output_price: number
+  cache_read_price: number
+  enabled: boolean
+  sort: number
+  channels?: ModelBinding[]
+  used?: boolean
+  needs_pricing?: boolean
+  denied?: boolean
+  created_at?: string
+  updated_at?: string
+  [key: string]: unknown
+}
+
+export type ModelBinding = {
+  id: number
+  channel_id: number
+  channel_name: string
+  upstream_model: string
+  weight: number
+}
+
+export type NewModelPayload = {
+  name: string
+  display_name?: string
+  input_price?: number
+  output_price?: number
+  cache_read_price?: number
+  sort?: number
+  enabled?: boolean
+}
+
+export type ModelSummary = {
+  total: number
+  unpriced: number
+  missing: OrphanBinding[]
+  denied_count: number
+}
+
+export type OrphanBinding = {
+  channel: string
+  model_id: number
+  upstream_model: string
+}
+
+export const useModelStore = defineStore('model', () => {
+  const loading = ref(false);
+  const error = ref<string | null>(null);
+  const models = ref<Model[]>([]);
+  const summary = ref<ModelSummary | null>(null);
+
+  const fetchModels = async () => {
+    loading.value = true;
+    error.value = null;
+    try {
+      const response = await request.get('/admin/models');
+      models.value = response.data.data ?? [];
+      summary.value = response.data.summary ?? null;
+    } catch (err: any) {
+      error.value = err.response?.data?.error || 'Failed to fetch models';
+      throw err;
+    } finally {
+      loading.value = false;
+    }
+  };
+
+  const createModel = async (data: NewModelPayload) => {
+    loading.value = true;
+    error.value = null;
+    try {
+      const response: AxiosResponse = await request.post('/admin/models', data);
+      return response;
+    } catch (err: any) {
+      error.value = err.response?.data?.error || 'Failed to create model';
+      throw err;
+    } finally {
+      loading.value = false;
+    }
+  };
+
+  const updateModel = async (id: number | string, data: Partial<Model>) => {
+    loading.value = true;
+    error.value = null;
+    try {
+      const response: AxiosResponse = await request.put(`/admin/models/${id}`, data);
+      return response;
+    } catch (err: any) {
+      error.value = err.response?.data?.error || 'Failed to update model';
+      throw err;
+    } finally {
+      loading.value = false;
+    }
+  };
+
+  const deleteModel = async (id: number | string) => {
+    loading.value = true;
+    error.value = null;
+    try {
+      const response: AxiosResponse = await request.delete(`/admin/models/${id}`);
+      return response;
+    } catch (err: any) {
+      error.value = err.response?.data?.error || 'Failed to delete model';
+      throw err;
+    } finally {
+      loading.value = false;
+    }
+  };
+
+  const deleteUnusedModels = async () => {
+    loading.value = true;
+    error.value = null;
+    try {
+      const response: AxiosResponse = await request.delete('/admin/models/unused');
+      return response;
+    } catch (err: any) {
+      error.value = err.response?.data?.error || 'Failed to delete unused models';
+      throw err;
+    } finally {
+      loading.value = false;
+    }
+  };
+
+  const createModelBinding = async (modelId: number | string, data: { channel_id: number; upstream_model: string; weight?: number }) => {
+    loading.value = true;
+    error.value = null;
+    try {
+      const response: AxiosResponse = await request.post(`/admin/models/${modelId}/bindings`, data);
+      return response;
+    } catch (err: any) {
+      error.value = err.response?.data?.error || 'Failed to create binding';
+      throw err;
+    } finally {
+      loading.value = false;
+    }
+  };
+
+  const deleteModelBinding = async (modelId: number | string, bindingId: number | string) => {
+    loading.value = true;
+    error.value = null;
+    try {
+      const response: AxiosResponse = await request.delete(`/admin/models/${modelId}/bindings/${bindingId}`);
+      return response;
+    } catch (err: any) {
+      error.value = err.response?.data?.error || 'Failed to delete binding';
+      throw err;
+    } finally {
+      loading.value = false;
+    }
+  };
+
+  return {
+    loading, error,
+    models, summary,
+    fetchModels,
+    createModel,
+    updateModel,
+    deleteModel,
+    deleteUnusedModels,
+    createModelBinding,
+    deleteModelBinding,
+  };
+});
diff --git a/frontend/src/types/index.ts b/frontend/src/types/index.ts
index 3fbf1eb..b2476a2 100644
--- a/frontend/src/types/index.ts
+++ b/frontend/src/types/index.ts
@@ -115,3 +115,59 @@ export type NewUserPayload = {
   unlimited_quota?: boolean
   language?: string
 }
+
+// Channel 渠道管理
+export interface Channel {
+  id: number
+  name: string
+  provider: 'openai' | 'anthropic' | 'compatible'
+  formats: string[] // chat | responses | messages
+  base_url: string
+  base_urls?: Record<string, string> | null
+  api_key_masked: string
+  weight: number
+  priority: number
+  timeout_ms: number
+  max_concurrency: number
+  health_status: string
+  enabled: boolean
+  created_at: string
+}
+
+export interface ChannelModelMapping {
+  id: number
+  model_id: number
+  model_name: string
+  upstream_model: string
+  weight: number
+}
+
+// Model 模型定价
+export interface ModelBinding {
+  id: number
+  channel_id: number
+  channel_name: string
+  upstream_model: string
+  weight: number
+}
+
+export interface Model {
+  id: number
+  name: string
+  input_price: number
+  output_price: number
+  cache_read_price: number
+  enabled: boolean
+  sort: number
+  channels: ModelBinding[]
+  used?: boolean
+  needs_pricing?: boolean
+  denied?: boolean
+}
+
+export interface ModelSummary {
+  total: number
+  unpriced: number
+  missing: { channel: string; model_id: number; upstream_model: string }[]
+  denied_count: number
+}
diff --git a/frontend/src/utils/router_menu.ts b/frontend/src/utils/router_menu.ts
index eab1495..104c0da 100644
--- a/frontend/src/utils/router_menu.ts
+++ b/frontend/src/utils/router_menu.ts
@@ -6,6 +6,8 @@ import {
   KeyRoundIcon,
   SettingsIcon,
   GlobeIcon,
+  BoxesIcon,
+  SlidersHorizontalIcon,
 } from '@lucide/vue'
 
 export type MenuLink = { label: string; to: string; icon?: Component }
@@ -48,8 +50,9 @@ export const routes: RouteRecordRaw[] = [
           { path: 'users', name: 'User', component: () => import('@/views/dashboard/User.vue'), meta: { title: '用户管理' } },
           { path: 'users/new', name: 'UserNew', component: () => import('@/views/dashboard/UserNew.vue'), meta: { title: '新建用户' } },
           { path: 'users/view', name: 'UserView', component: () => import('@/views/dashboard/UserView.vue'), meta: { title: '用户详情' } },
-          { path: 'channels', name: 'Channels', component: () => import('@/views/dashboard/Keys.vue'), meta: { title: '渠道管理' } },
-          { path: 'channels/view', name: 'ChannelView', component: () => import('@/views/dashboard/KeyView.vue'), meta: { title: '渠道详情' } },
+          { path: 'channels', name: 'Channels', component: () => import('@/views/dashboard/ChannelsView.vue'), meta: { title: '渠道管理' } },
+          { path: 'models', name: 'Models', component: () => import('@/views/dashboard/Models.vue'), meta: { title: '模型定价' } },
+          { path: 'config', name: 'SystemConfig', component: () => import('@/views/dashboard/SystemConfig.vue'), meta: { title: '系统配置' } },
         ],
       },
       {
@@ -76,4 +79,6 @@ export const consoleMenu: MenuLink[] = [
 export const adminMenu: MenuLink[] = [
   { label: '用户管理', to: '/dashboard/manager/users', icon: UsersRoundIcon },
   { label: '渠道管理', to: '/dashboard/manager/channels', icon: GlobeIcon },
+  { label: '模型定价', to: '/dashboard/manager/models', icon: BoxesIcon },
+  { label: '系统配置', to: '/dashboard/manager/config', icon: SlidersHorizontalIcon },
 ]
diff --git a/frontend/src/views/dashboard/ChannelModelsDrawer.vue b/frontend/src/views/dashboard/ChannelModelsDrawer.vue
new file mode 100644
index 0000000..274dcf7
--- /dev/null
+++ b/frontend/src/views/dashboard/ChannelModelsDrawer.vue
@@ -0,0 +1,183 @@
+<script setup lang="ts">
+import { onMounted, reactive, ref } from 'vue'
+import { RefreshCw, Plus, X } from '@lucide/vue'
+import request from '@/api/client'
+import { useToast } from '@/composables/toast'
+import Button from '@/components/ui/Button.vue'
+import type { Channel, ChannelModelMapping } from '@/types'
+
+function errMsg(e: unknown) {
+  return (e as any)?.response?.data?.error || (e as any)?.message || '请求失败'
+}
+
+const props = defineProps<{ channel: Channel }>()
+const { setToast } = useToast()
+
+const mappings = ref<ChannelModelMapping[]>([])
+const remote = ref<string[]>([])
+const selected = ref<string[]>([])
+const loading = ref(false)
+const fetched = ref(false)
+const addForm = reactive({ custom_name: '', upstream_model: '' })
+
+async function load() {
+  try {
+    const { data } = await request.get(`/admin/channels/${props.channel.id}/models`)
+    mappings.value = data.data?.items || data.data || []
+  } catch (e) {
+    setToast(errMsg(e), 'error')
+  }
+}
+
+async function fetchRemote() {
+  loading.value = true
+  try {
+    const { data } = await request.get(`/admin/channels/${props.channel.id}/models/remote`)
+    remote.value = data.data || []
+    selected.value = []
+    fetched.value = true
+  } catch (e) {
+    setToast(errMsg(e), 'error')
+  } finally {
+    loading.value = false
+  }
+}
+
+async function addSelected() {
+  let added = 0
+  for (const name of selected.value) {
+    try {
+      await request.post(`/admin/channels/${props.channel.id}/models`, {
+        upstream_model: name,
+      })
+      added++
+    } catch {
+      /* 单个失败不中断 */
+    }
+  }
+  selected.value = []
+  setToast(added ? `已添加 ${added} 个模型` : '所选均已添加', 'success')
+  await load()
+  await fetchRemote()
+}
+
+async function addManual() {
+  if (!addForm.upstream_model.trim()) return
+  try {
+    await request.post(`/admin/channels/${props.channel.id}/models`, {
+      upstream_model: addForm.upstream_model.trim(),
+      custom_name: addForm.custom_name.trim() || undefined,
+    })
+    setToast('已添加', 'success')
+    addForm.custom_name = ''
+    addForm.upstream_model = ''
+    await load()
+  } catch (e) {
+    setToast(errMsg(e), 'error')
+  }
+}
+
+async function saveUpstream(b: ChannelModelMapping) {
+  try {
+    await request.patch(`/admin/channels/${props.channel.id}/models/${b.id}`, {
+      upstream_model: b.upstream_model,
+    })
+    setToast('已更新', 'success')
+    await load()
+  } catch (e) {
+    setToast(errMsg(e), 'error')
+  }
+}
+
+async function remove(b: ChannelModelMapping) {
+  if (!confirm(`解除模型 ${b.model_name} 的绑定？`)) return
+  try {
+    await request.delete(`/admin/channels/${props.channel.id}/models/${b.id}`)
+    setToast('已解除', 'success')
+    await load()
+  } catch (e) {
+    setToast(errMsg(e), 'error')
+  }
+}
+
+onMounted(load)
+</script>
+
+<template>
+  <div class="space-y-3">
+    <!-- 已允许的模型 -->
+    <div>
+      <p class="mb-1.5 text-xs font-medium text-base-content/50">已允许的模型（{{ mappings.length }}）</p>
+      <div v-if="mappings.length" class="flex flex-wrap gap-2">
+        <div
+          v-for="b in mappings"
+          :key="b.id"
+          class="inline-flex items-center gap-1.5 rounded-md border border-base-300/60 bg-base-100 px-2 py-1 font-mono text-[11px] text-base-content/60"
+        >
+          <span class="text-base-content">{{ b.model_name }}</span>
+          <span class="opacity-60">→</span>
+          <input
+            v-model="b.upstream_model"
+            class="w-28 rounded border border-transparent bg-transparent px-1 text-[11px] text-primary outline-none transition focus:border-primary/50 focus:bg-base-200/50"
+            @change="saveUpstream(b)"
+          />
+          <button class="text-base-content/40 hover:text-error" aria-label="解除" @click="remove(b)">
+            <X :size="12" />
+          </button>
+        </div>
+      </div>
+      <p v-else class="text-xs text-base-content/50">尚未允许任何模型</p>
+    </div>
+
+    <!-- 从接口拉取 + 勾选 -->
+    <div class="border-t border-base-300/60 pt-3">
+      <div class="mb-1.5 flex items-center justify-between">
+        <p class="text-xs font-medium text-base-content/50">从接口拉取模型</p>
+        <Button size="sm" variant="ghost" :loading="loading" @click="fetchRemote">
+          <RefreshCw :size="13" />
+          拉取
+        </Button>
+      </div>
+      <div v-if="remote.length" class="flex max-h-36 flex-wrap gap-2 overflow-y-auto">
+        <label
+          v-for="m in remote"
+          :key="m"
+          class="flex cursor-pointer items-center gap-1.5 rounded-md border px-2 py-1 font-mono text-[11px] text-base-content/60 transition select-none"
+          :class="selected.includes(m) ? 'border-primary bg-primary/10 text-base-content' : 'border-base-300/60 hover:border-base-content/30'"
+        >
+          <input v-model="selected" type="checkbox" :value="m" class="size-3.5 accent-primary" />
+          {{ m }}
+        </label>
+      </div>
+      <div v-if="remote.length" class="mt-2">
+        <Button size="sm" @click="addSelected">
+          <Plus :size="13" />
+          添加所选（{{ selected.length }}）
+        </Button>
+      </div>
+      <p v-else-if="!loading" class="text-xs text-base-content/50">
+        {{ remote.length === 0 && fetched ? '接口返回的模型均已允许，无新增候选' : '点「拉取」获取渠道接口返回的新模型，勾选需要的加入' }}
+      </p>
+    </div>
+
+    <!-- 手动添加 -->
+    <div class="flex items-center gap-2 border-t border-base-300/60 pt-3">
+      <input
+        v-model="addForm.custom_name"
+        placeholder="自定义名称（可选）"
+        class="h-8 min-w-0 flex-1 rounded-md border border-base-300/60 bg-base-100 px-2 font-mono text-xs outline-none focus:border-primary"
+        @keyup.enter="addManual"
+      />
+      <input
+        v-model="addForm.upstream_model"
+        placeholder="上游模型名"
+        class="h-8 min-w-0 flex-1 rounded-md border border-base-300/60 bg-base-100 px-2 font-mono text-xs outline-none focus:border-primary"
+        @keyup.enter="addManual"
+      />
+      <Button size="sm" class="shrink-0" @click="addManual">
+        <Plus :size="13" />
+        添加
+      </Button>
+    </div>
+  </div>
+</template>
diff --git a/frontend/src/views/dashboard/ChannelsView.vue b/frontend/src/views/dashboard/ChannelsView.vue
new file mode 100644
index 0000000..c2d8aef
--- /dev/null
+++ b/frontend/src/views/dashboard/ChannelsView.vue
@@ -0,0 +1,342 @@
+<script setup lang="ts">
+import { onMounted, reactive, ref } from 'vue'
+import { ChevronDown, Zap, Pencil, Trash2, Layers } from '@lucide/vue'
+import request from '@/api/client'
+import { useToast } from '@/composables/toast'
+import { PROTOCOL_OPTIONS, protocolShort } from '@/lib/protocol'
+
+function errMsg(e: unknown) {
+  return (e as any)?.response?.data?.error || (e as any)?.message || '请求失败'
+}
+import ChannelModelsDrawer from '@/views/dashboard/ChannelModelsDrawer.vue'
+import Button from '@/components/ui/Button.vue'
+import Input from '@/components/ui/Input.vue'
+import Modal from '@/components/ui/Modal.vue'
+import Badge from '@/components/ui/Badge.vue'
+import type { Channel } from '@/types'
+
+const { setToast } = useToast()
+const channels = ref<Channel[]>([])
+const editOpen = ref(false)
+const editing = ref<Channel | null>(null)
+const saving = ref(false)
+const busyId = ref<number | null>(null)
+const expandedId = ref<number | null>(null)
+
+function toggleDrawer(ch: Channel) {
+  expandedId.value = expandedId.value === ch.id ? null : ch.id
+}
+
+const form = reactive({
+  name: '',
+  formats: ['chat'] as string[],
+  base_url: '',
+  base_urls: { chat: '', responses: '', messages: '' } as Record<string, string>,
+  api_key: '',
+  weight: 1,
+  priority: 0,
+  timeout_ms: 120000,
+  max_concurrency: 16,
+  enabled: true,
+})
+
+async function load() {
+  try {
+    const { data } = await request.get('/admin/channels')
+    channels.value = data.data.items || data.data
+  } catch (e) {
+    setToast(errMsg(e), 'error')
+  }
+}
+
+function openCreate() {
+  editing.value = null
+  Object.assign(form, {
+    name: '', formats: ['chat'], base_url: '',
+    base_urls: { chat: '', responses: '', messages: '' },
+    api_key: '',
+    weight: 1, priority: 0, timeout_ms: 120000, max_concurrency: 16, enabled: true,
+  })
+  editOpen.value = true
+}
+
+function openEdit(ch: Channel) {
+  editing.value = ch
+  Object.assign(form, {
+    name: ch.name, formats: [...(ch.formats?.length ? ch.formats : ['chat'])],
+    base_url: ch.base_url,
+    base_urls: {
+      chat: ch.base_urls?.chat ?? '',
+      responses: ch.base_urls?.responses ?? '',
+      messages: ch.base_urls?.messages ?? '',
+    },
+    api_key: '',
+    weight: ch.weight, priority: ch.priority, timeout_ms: ch.timeout_ms,
+    max_concurrency: ch.max_concurrency, enabled: ch.enabled,
+  })
+  editOpen.value = true
+}
+
+async function save() {
+  if (form.formats.length === 0) {
+    setToast('请至少选择一种 API 格式', 'error')
+    return
+  }
+  saving.value = true
+  const payload = {
+    ...form,
+    weight: Number(form.weight),
+    priority: Number(form.priority),
+    timeout_ms: Number(form.timeout_ms),
+    max_concurrency: Number(form.max_concurrency),
+  }
+  try {
+    if (editing.value) {
+      await request.put(`/admin/channels/${editing.value.id}`, payload)
+      setToast('渠道已更新', 'success')
+    } else {
+      await request.post('/admin/channels', payload)
+      setToast('渠道已创建', 'success')
+    }
+    editOpen.value = false
+    await load()
+  } catch (e) {
+    setToast(errMsg(e), 'error')
+  } finally {
+    saving.value = false
+  }
+}
+
+async function remove(ch: Channel) {
+  if (!confirm(`删除渠道 ${ch.name}？关联的模型绑定也会清除。`)) return
+  try {
+    await request.delete(`/admin/channels/${ch.id}`)
+    setToast('渠道已删除', 'success')
+    await load()
+  } catch (e) {
+    setToast(errMsg(e), 'error')
+  }
+}
+
+async function testChannel(ch: Channel) {
+  busyId.value = ch.id
+  try {
+    await request.post(`/admin/channels/${ch.id}/test`)
+    setToast(`渠道 ${ch.name} 连接正常`, 'success')
+  } catch (e) {
+    setToast(`连接失败: ${errMsg(e)}`, 'error')
+  } finally {
+    busyId.value = null
+    await load()
+  }
+}
+
+onMounted(load)
+</script>
+
+<template>
+  <div class="mx-auto max-w-6xl">
+    <div class="mb-6 flex flex-wrap items-center justify-between gap-3">
+      <div>
+        <h1 class="text-lg font-semibold">渠道</h1>
+        <p class="text-sm text-base-content/60">接入上游服务，API Key 加密存储</p>
+      </div>
+      <Button class="shrink-0" @click="openCreate">添加渠道</Button>
+    </div>
+
+    <!-- 移动端：卡片列表 -->
+    <div class="space-y-3 md:hidden">
+      <div v-for="ch in channels" :key="ch.id" class="card border border-base-300/60 bg-base-100 p-4 shadow-sm" :class="ch.enabled ? 'border-l-2 border-l-success' : ''">
+        <div class="flex flex-wrap items-start justify-between gap-2">
+          <div class="min-w-0">
+            <p class="text-sm font-medium">{{ ch.name }}</p>
+            <div class="mt-1.5 flex flex-wrap gap-1">
+              <code
+                v-for="f in ch.formats || []"
+                :key="f"
+                class="rounded bg-base-200 px-1.5 py-0.5 font-mono text-[10px] text-base-content/60"
+              >{{ protocolShort(f) }}</code>
+            </div>
+          </div>
+          <div class="flex shrink-0 gap-1.5">
+            <Badge :variant="ch.health_status === 'healthy' ? 'ok' : ch.health_status === 'cooldown' ? 'err' : 'warn'">
+              {{ ch.health_status }}
+            </Badge>
+            <Badge :variant="ch.enabled ? 'ok' : 'neutral'">{{ ch.enabled ? '启用' : '停用' }}</Badge>
+          </div>
+        </div>
+        <p class="mt-2 truncate font-mono text-[11px] text-base-content/60">{{ ch.base_url }}</p>
+        <div class="mt-3 flex flex-wrap gap-x-3 gap-y-1.5 border-t border-base-300/60 pt-3">
+          <button class="inline-flex items-center gap-1 text-xs text-base-content/60 hover:text-primary" :disabled="busyId === ch.id" @click="testChannel(ch)">
+            <Zap :size="13" />
+            {{ busyId === ch.id ? '测试中…' : '测试' }}
+          </button>
+          <button class="inline-flex items-center gap-1 text-xs text-primary hover:text-primary/80" @click="toggleDrawer(ch)">
+            <Layers :size="13" />
+            支持的模型 {{ expandedId === ch.id ? '▴' : '▾' }}
+          </button>
+          <button class="inline-flex items-center gap-1 text-xs text-base-content/60 hover:text-base-content" @click="openEdit(ch)">
+            <Pencil :size="13" />
+            编辑
+          </button>
+          <button class="inline-flex items-center gap-1 text-xs text-base-content/60 hover:text-error" @click="remove(ch)">
+            <Trash2 :size="13" />
+            删除
+          </button>
+        </div>
+        <div v-if="expandedId === ch.id" class="mt-3 border-t border-base-300/60 pt-3">
+          <ChannelModelsDrawer :channel="ch" />
+        </div>
+      </div>
+      <p v-if="channels.length === 0" class="card border border-base-300/60 bg-base-100 px-4 py-10 text-center text-sm text-base-content/60">
+        还没有渠道，点击「添加渠道」
+      </p>
+    </div>
+
+    <!-- 桌面端：表格 -->
+    <div class="card hidden border border-base-300/60 bg-base-100 shadow-sm md:block">
+      <div class="overflow-x-auto">
+        <table class="w-full text-sm min-w-[820px]">
+          <thead>
+            <tr class="border-b border-base-300/60 text-left text-xs text-base-content/50">
+              <th scope="col" class="px-4 py-2.5 font-medium">名称</th>
+              <th scope="col" class="px-4 py-2.5 font-medium">API 格式</th>
+              <th scope="col" class="px-4 py-2.5 font-medium">Base URL</th>
+              <th scope="col" class="px-4 py-2.5 font-medium">Key</th>
+              <th scope="col" class="px-4 py-2.5 font-medium">健康</th>
+              <th scope="col" class="px-4 py-2.5 font-medium">启用</th>
+              <th scope="col" class="px-4 py-2.5" />
+            </tr>
+          </thead>
+          <tbody>
+            <template v-for="ch in channels" :key="ch.id">
+            <tr class="border-b border-base-300/40 last:border-0 hover:bg-base-200/50" :style="ch.enabled ? { borderLeft: '2px solid oklch(var(--p))' } : {}">
+              <td class="px-4 py-2.5">
+                <button class="inline-flex items-center gap-1.5 transition hover:text-primary" @click="toggleDrawer(ch)">
+                  <span class="truncate">{{ ch.name }}</span>
+                  <ChevronDown :size="12" class="shrink-0 text-base-content/50 transition-transform" :class="expandedId === ch.id ? 'rotate-180' : ''" />
+                </button>
+              </td>
+              <td class="px-4 py-2.5">
+                <div class="flex flex-col gap-0.5">
+                  <code
+                    v-for="f in ch.formats || []"
+                    :key="f"
+                    class="font-mono text-[11px] leading-4 text-base-content/60"
+                  >{{ protocolShort(f) }}</code>
+                </div>
+              </td>
+              <td class="max-w-[220px] truncate px-4 py-2.5 font-mono text-xs text-base-content/60">{{ ch.base_url }}</td>
+              <td class="px-4 py-2.5 font-mono text-xs text-base-content/60">{{ ch.api_key_masked || '****' }}</td>
+              <td class="px-4 py-2.5">
+                <Badge :variant="ch.health_status === 'healthy' ? 'ok' : ch.health_status === 'cooldown' ? 'err' : 'warn'">
+                  {{ ch.health_status }}
+                </Badge>
+              </td>
+              <td class="px-4 py-2.5 text-xs text-base-content/60">{{ ch.enabled ? '是' : '否' }}</td>
+              <td class="px-4 py-2.5 text-right">
+                <div class="flex justify-end gap-2">
+                  <button class="inline-flex items-center gap-1 text-xs text-base-content/60 hover:text-primary" :disabled="busyId === ch.id" @click="testChannel(ch)">
+                    <Zap :size="13" />
+                    {{ busyId === ch.id ? '测试中…' : '测试' }}
+                  </button>
+                  <button class="inline-flex items-center gap-1 text-xs text-base-content/60 hover:text-base-content" @click="openEdit(ch)">
+                    <Pencil :size="13" />
+                    编辑
+                  </button>
+                  <button class="inline-flex items-center gap-1 text-xs text-base-content/60 hover:text-error" @click="remove(ch)">
+                    <Trash2 :size="13" />
+                    删除
+                  </button>
+                </div>
+              </td>
+            </tr>
+            <tr v-if="expandedId === ch.id" class="bg-base-200/30">
+              <td colspan="7" class="px-4 py-3">
+                <ChannelModelsDrawer :channel="ch" />
+              </td>
+            </tr>
+            </template>
+            <tr v-if="channels.length === 0">
+              <td colspan="7" class="px-4 py-10 text-center text-sm text-base-content/60">还没有渠道，点击「添加渠道」</td>
+            </tr>
+          </tbody>
+        </table>
+      </div>
+    </div>
+
+    <Modal :open="editOpen" :title="editing ? '编辑渠道' : '添加渠道'" @close="editOpen = false">
+      <div class="space-y-4">
+        <Input v-model="form.name" label="名称" placeholder="openai" />
+        <div>
+          <span class="mb-1.5 block text-xs font-medium text-base-content/50">支持的 API 格式</span>
+          <div class="flex flex-wrap gap-2">
+            <label
+              v-for="opt in PROTOCOL_OPTIONS"
+              :key="opt.value"
+              class="flex cursor-pointer items-center gap-1.5 rounded-md border px-2.5 py-1.5 text-xs transition select-none"
+              :class="form.formats.includes(opt.value) ? 'border-primary bg-primary/10' : 'border-base-300 text-base-content/60 hover:border-base-content/30'"
+            >
+              <input
+                v-model="form.formats"
+                type="checkbox"
+                :value="opt.value"
+                class="size-3.5 accent-primary"
+              />
+              {{ opt.label }}
+            </label>
+          </div>
+          <p class="mt-1.5 text-xs text-base-content/50">客户端协议不在其中时，网关自动转换为其支持的格式</p>
+        </div>
+        <Input
+          v-model="form.base_url"
+          label="Base URL（可选）"
+          placeholder="https://api.openai.com/v1"
+          :maxlength="255"
+          hint="支持前缀或完整端点，如 https://api.openai.com/v1 或 https://api.openai.com/v1/chat/completions；留空按供应商默认"
+        />
+        <div class="space-y-3 rounded-md border border-base-300/60 p-3">
+          <p class="text-xs font-medium text-base-content/50">分协议 Base URL（可选，如智谱三种格式不同）</p>
+          <Input v-model="form.base_urls.chat" label="OpenAI Chat Completions" placeholder="留空用主 Base URL" :maxlength="255" />
+          <Input v-model="form.base_urls.responses" label="OpenAI Responses" placeholder="留空用主 Base URL" :maxlength="255" />
+          <Input v-model="form.base_urls.messages" label="Anthropic Messages" placeholder="留空用主 Base URL" :maxlength="255" />
+          <p class="text-xs text-base-content/50">网关按协议选对应 base_url 直通，无需为每种格式建多个渠道</p>
+        </div>
+        <Input
+          v-model="form.api_key"
+          label="上游 API Key"
+          :placeholder="editing ? '留空则不修改' : 'sk-...'"
+        />
+        <div class="grid grid-cols-1 gap-4 sm:grid-cols-2">
+          <Input v-model="form.weight" label="权重" type="number" />
+          <Input v-model="form.priority" label="优先级" type="number" />
+          <Input v-model="form.timeout_ms" label="超时 (ms)" type="number" />
+          <Input v-model="form.max_concurrency" label="最大并发" type="number" />
+        </div>
+        <div class="flex items-center justify-between rounded-md border border-base-300/60 p-3">
+          <div>
+            <p class="text-sm font-medium">启用渠道</p>
+            <p class="text-xs text-base-content/50">禁用后该渠道不会被用于请求转发</p>
+          </div>
+          <button
+            type="button"
+            role="switch"
+            :aria-checked="form.enabled"
+            class="relative inline-flex h-6 w-11 shrink-0 cursor-pointer items-center rounded-full transition-colors focus-visible:outline-2 focus-visible:outline-offset-2 focus-visible:outline-primary"
+            :class="form.enabled ? 'bg-primary' : 'bg-base-200'"
+            @click="form.enabled = !form.enabled"
+          >
+            <span
+              class="pointer-events-none inline-block h-4 w-4 rounded-full bg-white shadow-sm ring-0 transition-transform"
+              :class="form.enabled ? 'translate-x-6' : 'translate-x-1'"
+            />
+          </button>
+        </div>
+      </div>
+      <template #footer>
+        <Button variant="ghost" @click="editOpen = false">取消</Button>
+        <Button :loading="saving" @click="save">{{ editing ? '保存' : '创建' }}</Button>
+      </template>
+    </Modal>
+  </div>
+</template>
diff --git a/frontend/src/views/dashboard/KeyView.vue b/frontend/src/views/dashboard/KeyView.vue
index af0d71d..adce801 100644
--- a/frontend/src/views/dashboard/KeyView.vue
+++ b/frontend/src/views/dashboard/KeyView.vue
@@ -87,6 +87,10 @@
 
           <div class="flex items-center justify-end gap-3 border-t border-base-300/40 pt-4">
             <button type="button" @click="goBack" class="btn btn-ghost btn-sm">Back</button>
+            <button type="button" @click="testChannel" class="btn btn-warning btn-sm" :disabled="testing">
+              <span v-if="testing" class="loading loading-spinner loading-xs" aria-hidden="true"></span>
+              Test Connection
+            </button>
             <button type="submit" class="btn btn-primary btn-sm px-5" :disabled="updating">
               <span v-if="updating" class="loading loading-spinner loading-xs" aria-hidden="true"></span>
               Save Changes
@@ -94,6 +98,47 @@
           </div>
         </form>
       </div>
+
+      <!-- Model Bindings -->
+      <div class="card border border-base-300/60 bg-base-100 shadow-sm">
+        <div class="card-body gap-4 p-4 sm:p-6">
+          <div class="flex items-center justify-between">
+            <h2 class="text-xs font-semibold uppercase tracking-wider text-base-content/50">Model Bindings</h2>
+            <button class="btn btn-primary btn-sm" @click="openAddModelModal">
+              <PlusIcon class="h-4 w-4" aria-hidden="true" />Add Model
+            </button>
+          </div>
+          
+          <div v-if="bindings.length > 0" class="overflow-x-auto">
+            <table class="table table-sm">
+              <thead>
+                <tr class="text-xs uppercase tracking-wider text-base-content/50">
+                  <th>Model Name</th>
+                  <th>Upstream Model</th>
+                  <th class="text-right">Weight</th>
+                  <th class="text-right"><span class="sr-only">Actions</span></th>
+                </tr>
+              </thead>
+              <tbody>
+                <tr v-for="b in bindings" :key="b.id" class="border-base-300/40">
+                  <td class="font-medium">{{ b.model_name }}</td>
+                  <td class="font-mono text-xs">{{ b.upstream_model }}</td>
+                  <td class="text-right">{{ b.weight }}</td>
+                  <td class="text-right">
+                    <button class="btn btn-ghost btn-xs btn-square text-error" @click="confirmDeleteBinding(b)"
+                      aria-label="Delete binding">
+                      <TrashIcon class="h-4 w-4" aria-hidden="true" />
+                    </button>
+                  </td>
+                </tr>
+              </tbody>
+            </table>
+          </div>
+          <div v-else class="py-6 text-center text-sm text-base-content/50">
+            No model bindings configured.
+          </div>
+        </div>
+      </div>
     </div>
 
     <!-- Loading state -->
@@ -104,22 +149,59 @@
         </div>
       </div>
     </div>
+
+    <!-- Add Model Modal -->
+    <dialog ref="addModelModalRef" class="modal">
+      <div class="modal-box max-w-lg px-0 sm:px-6">
+        <form method="dialog">
+          <button class="btn btn-circle btn-ghost btn-sm absolute right-2 top-2" aria-label="Close dialog">✕</button>
+        </form>
+        <h3 class="mb-4 text-lg font-bold">Add Model Binding</h3>
+        <form @submit.prevent="addModelBinding" class="space-y-4">
+          <label class="floating-label">
+            <span>Model ID *</span>
+            <input v-model.number="newBinding.model_id" type="number" placeholder="Model ID" class="input w-full" required />
+          </label>
+          <label class="floating-label">
+            <span>Upstream Model Name *</span>
+            <input v-model="newBinding.upstream_model" type="text" placeholder="e.g. gpt-4o" class="input w-full" required />
+          </label>
+          <label class="floating-label">
+            <span>Weight</span>
+            <input v-model.number="newBinding.weight" type="number" min="1" placeholder="1" class="input w-full" />
+          </label>
+          <div class="modal-action">
+            <button type="button" class="btn btn-ghost" @click="closeAddModelModal">Cancel</button>
+            <button type="submit" class="btn btn-primary" :disabled="addingModel">
+              {{ addingModel ? 'Adding...' : 'Add' }}
+            </button>
+          </div>
+        </form>
+      </div>
+      <form method="dialog" class="modal-backdrop">
+        <button aria-label="Close dialog">close</button>
+      </form>
+    </dialog>
   </div>
 </template>
 
 <script setup lang="ts">
 import { computed, onMounted, ref } from 'vue';
 import { useRoute, useRouter } from 'vue-router';
-import { useChannelStore, type Channel } from '../../stores/channel';
+import { useChannelStore, type Channel, type ChannelModelBinding } from '../../stores/channel';
 import BreadcrumbHeader from '@/components/dashboard/BreadcrumbHeader.vue';
 import { useToast } from '@/composables/toast';
 
+import { PlusIcon, TrashIcon } from '@lucide/vue';
+
 const route = useRoute();
 const router = useRouter();
 const channelStore = useChannelStore();
 const { setToast } = useToast();
 const updating = ref(false);
+const testing = ref(false);
 const api_key = ref('');
+const bindings = ref<ChannelModelBinding[]>([]);
 
 const channelId = computed(() => route.query.id);
 const ch = computed(() => channelStore.channel);
@@ -127,9 +209,16 @@ const ch = computed(() => channelStore.channel);
 onMounted(async () => {
   if (channelId.value) {
     await channelStore.fetchChannel(channelId.value as string);
+    await fetchBindings();
   }
 });
 
+const fetchBindings = async () => {
+  if (channelId.value) {
+    bindings.value = await channelStore.fetchChannelModels(channelId.value as string);
+  }
+};
+
 const toggleEnabled = () => {
   if (!ch.value) return;
   ch.value.enabled = !ch.value.enabled;
@@ -163,7 +252,65 @@ const updateCh = async () => {
   }
 };
 
+const testChannel = async () => {
+  if (!ch.value) return;
+  testing.value = true;
+  try {
+    const result = await channelStore.testChannel(ch.value.id);
+    setToast(`Connection OK (${result.data?.latency_ms}ms)`, 'success');
+  } catch (err: any) {
+    setToast(err.response?.data?.error || 'Connection test failed', 'error');
+  } finally {
+    testing.value = false;
+  }
+};
+
 const goBack = () => {
   router.push({ name: 'Channels' });
 };
+
+// Model binding
+const addModelModalRef = ref<HTMLDialogElement | null>(null);
+const addingModel = ref(false);
+const newBinding = ref({
+  model_id: 0,
+  upstream_model: '',
+  weight: 1,
+});
+
+const openAddModelModal = () => {
+  newBinding.value = { model_id: 0, upstream_model: '', weight: 1 };
+  addModelModalRef.value?.showModal();
+};
+
+const closeAddModelModal = () => {
+  addModelModalRef.value?.close();
+};
+
+const addModelBinding = async () => {
+  if (!channelId.value || !newBinding.value.model_id || !newBinding.value.upstream_model) return;
+  addingModel.value = true;
+  try {
+    await channelStore.addChannelModel(channelId.value as string, newBinding.value);
+    setToast('Model binding added', 'success');
+    closeAddModelModal();
+    await fetchBindings();
+  } catch (err: any) {
+    setToast(err.response?.data?.error || 'Failed to add binding', 'error');
+  } finally {
+    addingModel.value = false;
+  }
+};
+
+const confirmDeleteBinding = async (b: ChannelModelBinding) => {
+  if (confirm(`Remove binding for model "${b.model_name}"?`)) {
+    try {
+      await channelStore.deleteChannelModel(channelId.value as string, b.id);
+      setToast('Binding removed', 'success');
+      await fetchBindings();
+    } catch (err: any) {
+      setToast('Delete failed', 'error');
+    }
+  }
+};
 </script>
diff --git a/frontend/src/views/dashboard/Models.vue b/frontend/src/views/dashboard/Models.vue
new file mode 100644
index 0000000..b3ab5a0
--- /dev/null
+++ b/frontend/src/views/dashboard/Models.vue
@@ -0,0 +1,277 @@
+<template>
+  <div class="space-y-5">
+    <BreadcrumbHeader />
+
+    <div class="flex flex-wrap items-center justify-between gap-3">
+      <div>
+        <p class="text-sm text-base-content/60">Manage model pricing and channel bindings.</p>
+        <div v-if="summary" class="mt-1 flex gap-3 text-xs text-base-content/50">
+          <span>Total: {{ summary.total }}</span>
+          <span v-if="summary.unpriced > 0" class="text-warning">{{ summary.unpriced }} unpriced</span>
+          <span v-if="summary.missing.length > 0" class="text-error">{{ summary.missing.length }} orphan bindings</span>
+        </div>
+      </div>
+      <div class="flex items-center gap-2">
+        <button v-if="models.length > 0" class="btn btn-ghost btn-sm" @click="confirmDeleteUnused">
+          <TrashIcon class="h-4 w-4" aria-hidden="true" />Clean Unused
+        </button>
+        <button class="btn btn-primary btn-sm" @click="openCreateModal" aria-label="Create new model">
+          <PlusIcon class="h-4 w-4" aria-hidden="true" />New Model
+        </button>
+      </div>
+    </div>
+
+    <!-- Model cards -->
+    <div class="space-y-3">
+      <div v-for="m in models" :key="m.id"
+        class="card border bg-base-100 shadow-sm"
+        :class="m.channels && m.channels.length > 0 ? 'border-base-300/60' : 'border-warning/60 bg-warning/5'">
+        <div class="px-4 py-3">
+          <div class="flex flex-wrap items-center justify-between gap-x-4 gap-y-2">
+            <div class="flex flex-wrap items-center gap-2">
+              <span class="font-mono text-sm font-medium text-base-content">{{ m.name }}</span>
+              <span v-if="m.display_name" class="text-xs text-base-content/50">{{ m.display_name }}</span>
+              <span v-if="m.channels && m.channels.length > 0" class="badge badge-xs badge-ghost">渠道允许</span>
+              <span v-else class="badge badge-xs bg-yellow-200 text-yellow-800 dark:bg-yellow-900/50 dark:text-yellow-300">悬空</span>
+              <span :class="m.enabled ? 'badge badge-xs bg-green-200 text-green-800 dark:bg-green-900/50 dark:text-green-300' : 'badge badge-xs badge-ghost'">{{ m.enabled ? '启用' : '停用' }}</span>
+              <span v-if="m.denied" class="badge badge-xs badge-error">已禁止</span>
+              <span v-if="m.needs_pricing" class="badge badge-xs bg-orange-200 text-orange-800 dark:bg-orange-900/50 dark:text-orange-300">未定价</span>
+            </div>
+            <div class="flex gap-2">
+              <button class="btn btn-ghost btn-xs" @click="openEditModal(m)">编辑</button>
+              <button class="btn btn-ghost btn-xs text-error" @click="confirmDeleteModel(m)">删除</button>
+            </div>
+          </div>
+          <div class="mt-2 flex flex-wrap items-center gap-3">
+            <span class="font-mono text-xs text-base-content/60">入 {{ formatPrice(m.input_price) }}</span>
+            <span class="font-mono text-xs text-base-content/60">出 {{ formatPrice(m.output_price) }}</span>
+            <span class="font-mono text-xs text-base-content/60">缓存读 {{ formatPrice(m.cache_read_price) }}</span>
+          </div>
+        </div>
+
+        <div v-if="m.channels && m.channels.length > 0" class="border-t border-base-300/60 px-4 py-2">
+          <p class="mb-1.5 text-[11px] font-medium text-base-content/50">允许渠道</p>
+          <div class="flex flex-wrap gap-2">
+            <span v-for="ch in m.channels" :key="ch.id"
+              class="inline-flex items-center rounded-md border border-base-300/60 bg-base-100 px-2 py-0.5 font-mono text-[11px] text-base-content/60">
+              {{ ch.channel_name }} → {{ ch.upstream_model }}
+            </span>
+          </div>
+        </div>
+        <p v-else class="border-t border-base-300/60 bg-amber-100 px-4 py-2 text-xs font-medium text-amber-900 dark:bg-amber-900/40 dark:text-amber-100">
+          悬空模型：无任何渠道提供，客户端无法调用
+        </p>
+      </div>
+
+      <!-- Empty state -->
+      <div v-if="models.length === 0" class="card border border-base-300/60 bg-base-100 px-4 py-14 text-center">
+        <BoxesIcon class="mx-auto h-10 w-10 text-base-content/20" aria-hidden="true" />
+        <h2 class="mt-2 text-sm font-semibold">No models yet</h2>
+        <p class="mt-1 max-w-xs text-sm text-base-content/60">
+          Add models to manage pricing and channel bindings.
+        </p>
+        <button class="btn btn-primary btn-sm mt-3" @click="openCreateModal">
+          <PlusIcon class="h-4 w-4" aria-hidden="true" />Create Model
+        </button>
+      </div>
+    </div>
+
+    <!-- Create/Edit modal -->
+    <dialog ref="modalRef" class="modal">
+      <div class="modal-box max-w-lg px-0 sm:px-6">
+        <form method="dialog">
+          <button class="btn btn-circle btn-ghost btn-sm absolute right-2 top-2" aria-label="Close dialog">✕</button>
+        </form>
+        <h3 class="mb-4 text-lg font-bold">{{ editingModel ? 'Edit Model' : 'New Model' }}</h3>
+        <form @submit.prevent="saveModel" class="space-y-4">
+          <label class="floating-label">
+            <span>Model Name *</span>
+            <input v-model="form.name" type="text" placeholder="e.g. gpt-4o" class="input w-full" required
+              :disabled="!!editingModel" />
+          </label>
+          <label class="floating-label">
+            <span>Display Name</span>
+            <input v-model="form.display_name" type="text" placeholder="e.g. GPT-4o" class="input w-full" />
+          </label>
+          <div class="grid grid-cols-3 gap-3">
+            <label class="floating-label">
+              <span>Input $/M tokens</span>
+              <input v-model.number="form.input_price" type="number" step="0.01" min="0" placeholder="0"
+                class="input w-full" />
+            </label>
+            <label class="floating-label">
+              <span>Output $/M tokens</span>
+              <input v-model.number="form.output_price" type="number" step="0.01" min="0" placeholder="0"
+                class="input w-full" />
+            </label>
+            <label class="floating-label">
+              <span>Cache Read $/M</span>
+              <input v-model.number="form.cache_read_price" type="number" step="0.01" min="0" placeholder="0"
+                class="input w-full" />
+            </label>
+          </div>
+          <div class="grid grid-cols-2 gap-3">
+            <label class="floating-label">
+              <span>Sort Order</span>
+              <input v-model.number="form.sort" type="number" min="0" placeholder="0" class="input w-full" />
+            </label>
+            <div class="flex items-center gap-2 pt-6">
+              <input type="checkbox" class="toggle toggle-success toggle-sm" v-model="form.enabled" />
+              <span class="text-sm">{{ form.enabled ? 'Enabled' : 'Disabled' }}</span>
+            </div>
+          </div>
+          <div class="modal-action">
+            <button type="button" class="btn btn-ghost" @click="closeModal">Cancel</button>
+            <button type="submit" class="btn btn-primary" :disabled="saving">
+              {{ saving ? 'Saving...' : (editingModel ? 'Update' : 'Create') }}
+            </button>
+          </div>
+        </form>
+      </div>
+      <form method="dialog" class="modal-backdrop">
+        <button aria-label="Close dialog">close</button>
+      </form>
+    </dialog>
+  </div>
+</template>
+
+<script setup lang="ts">
+import { ref, reactive, onMounted } from 'vue';
+import BreadcrumbHeader from '@/components/dashboard/BreadcrumbHeader.vue';
+import { useModelStore, type Model, type NewModelPayload } from '@/stores/model';
+import { useToast } from '@/composables/toast';
+
+import {
+  BoxesIcon, PencilIcon, PlusIcon, TrashIcon
+} from '@lucide/vue';
+
+const modelStore = useModelStore();
+const { setToast } = useToast();
+
+const models = ref<Model[]>([]);
+const summary = ref(modelStore.summary);
+const editingModel = ref<Model | null>(null);
+const saving = ref(false);
+
+const form = reactive<NewModelPayload & { enabled: boolean }>({
+  name: '',
+  display_name: '',
+  input_price: 0,
+  output_price: 0,
+  cache_read_price: 0,
+  sort: 0,
+  enabled: true,
+});
+
+onMounted(async () => {
+  await fetchModels();
+});
+
+const fetchModels = async () => {
+  await modelStore.fetchModels();
+  models.value = modelStore.models;
+  summary.value = modelStore.summary;
+};
+
+const formatPrice = (price: number) => {
+  return price === 0 ? '-' : `$${price.toFixed(2)}`;
+};
+
+const toggleEnabled = async (m: Model) => {
+  try {
+    await modelStore.updateModel(m.id, { enabled: !m.enabled });
+    setToast(`Model ${m.name} ${m.enabled ? 'disabled' : 'enabled'}`, 'success');
+    await fetchModels();
+  } catch (error: any) {
+    setToast('Status update failed', 'error');
+  }
+};
+
+const openCreateModal = () => {
+  editingModel.value = null;
+  form.name = '';
+  form.display_name = '';
+  form.input_price = 0;
+  form.output_price = 0;
+  form.cache_read_price = 0;
+  form.sort = 0;
+  form.enabled = true;
+  modalRef.value?.showModal();
+};
+
+const openEditModal = (m: Model) => {
+  editingModel.value = m;
+  form.name = m.name;
+  form.display_name = m.display_name || '';
+  form.input_price = m.input_price;
+  form.output_price = m.output_price;
+  form.cache_read_price = m.cache_read_price;
+  form.sort = m.sort;
+  form.enabled = m.enabled;
+  modalRef.value?.showModal();
+};
+
+const saveModel = async () => {
+  saving.value = true;
+  try {
+    if (editingModel.value) {
+      await modelStore.updateModel(editingModel.value.id, {
+        display_name: form.display_name,
+        input_price: form.input_price,
+        output_price: form.output_price,
+        cache_read_price: form.cache_read_price,
+        sort: form.sort,
+        enabled: form.enabled,
+      });
+      setToast('Model updated', 'success');
+    } else {
+      await modelStore.createModel({
+        name: form.name,
+        display_name: form.display_name,
+        input_price: form.input_price,
+        output_price: form.output_price,
+        cache_read_price: form.cache_read_price,
+        sort: form.sort,
+        enabled: form.enabled,
+      });
+      setToast('Model created', 'success');
+    }
+    closeModal();
+    await fetchModels();
+  } catch (error: any) {
+    setToast(error.message || 'Save failed', 'error');
+  } finally {
+    saving.value = false;
+  }
+};
+
+const confirmDeleteModel = async (m: Model) => {
+  if (confirm(`Delete model "${m.name}"? This will also remove all channel bindings.`)) {
+    try {
+      await modelStore.deleteModel(m.id);
+      setToast(`Model ${m.name} deleted`, 'success');
+      await fetchModels();
+    } catch (error: any) {
+      setToast('Delete failed', 'error');
+    }
+  }
+};
+
+const confirmDeleteUnused = async () => {
+  if (confirm('Delete all models that are not bound to any channel?')) {
+    try {
+      const result = await modelStore.deleteUnusedModels();
+      const count = result.data?.count || 0;
+      setToast(`Deleted ${count} unused models`, 'success');
+      await fetchModels();
+    } catch (error: any) {
+      setToast('Cleanup failed', 'error');
+    }
+  }
+};
+
+const modalRef = ref<HTMLDialogElement | null>(null);
+const closeModal = () => {
+  modalRef.value?.close();
+};
+</script>
diff --git a/frontend/src/views/dashboard/ModelsView.vue b/frontend/src/views/dashboard/ModelsView.vue
new file mode 100644
index 0000000..8b76158
--- /dev/null
+++ b/frontend/src/views/dashboard/ModelsView.vue
@@ -0,0 +1,225 @@
+<script setup lang="ts">
+import { computed, onMounted, reactive, ref } from 'vue'
+import request from '@/api/client'
+import { useToast } from '@/composables/toast'
+import Button from '@/components/ui/Button.vue'
+import Input from '@/components/ui/Input.vue'
+import Modal from '@/components/ui/Modal.vue'
+import Badge from '@/components/ui/Badge.vue'
+import type { Model, ModelSummary } from '@/types'
+
+function errMsg(e: unknown) {
+  return (e as any)?.response?.data?.error || (e as any)?.message || '请求失败'
+}
+
+const { setToast } = useToast()
+const models = ref<Model[]>([])
+const summary = ref<ModelSummary>({ total: 0, unpriced: 0, missing: [], denied_count: 0 })
+const editOpen = ref(false)
+const editing = ref<Model | null>(null)
+const saving = ref(false)
+const quickName = ref('')
+const clearing = ref(false)
+
+const unused = computed(() => models.value.filter((m) => m.channels.length === 0))
+
+async function clearUnused() {
+  if (!unused.value.length) {
+    setToast('没有未绑定渠道的模型', 'info')
+    return
+  }
+  const names = unused.value.map((m) => m.name)
+  if (!confirm(`确定删除 ${names.length} 个未绑定渠道的模型？\n\n${names.join('\n')}`)) return
+  clearing.value = true
+  try {
+    const { data } = await request.delete('/admin/models/unused')
+    setToast(`已清除 ${data.data.count} 个模型`, 'success')
+    await load()
+  } catch (e) {
+    setToast(errMsg(e), 'error')
+  } finally {
+    clearing.value = false
+  }
+}
+
+function quickAdd() {
+  openCreate()
+  if (quickName.value) form.name = quickName.value.trim()
+}
+
+const form = reactive({
+  name: '',
+  input_price: 0,
+  output_price: 0,
+  cache_read_price: 0,
+  enabled: true,
+})
+
+async function load() {
+  try {
+    const { data } = await request.get('/admin/models')
+    models.value = data.data
+    summary.value = data.summary
+  } catch (e) {
+    setToast(errMsg(e), 'error')
+  }
+}
+
+function openCreate() {
+  editing.value = null
+  Object.assign(form, { name: '', input_price: 0, output_price: 0, cache_read_price: 0, enabled: true })
+  editOpen.value = true
+}
+
+function openEdit(m: Model) {
+  editing.value = m
+  Object.assign(form, {
+    name: m.name,
+    input_price: m.input_price, output_price: m.output_price, cache_read_price: m.cache_read_price,
+    enabled: m.enabled,
+  })
+  editOpen.value = true
+}
+
+async function save() {
+  saving.value = true
+  const payload = {
+    input_price: Number(form.input_price),
+    output_price: Number(form.output_price),
+    cache_read_price: Number(form.cache_read_price),
+    enabled: form.enabled,
+  }
+  try {
+    if (editing.value) {
+      await request.put(`/admin/models/${editing.value.id}`, payload)
+      setToast('模型已更新', 'success')
+    } else {
+      await request.post('/admin/models', { name: form.name, ...payload })
+      setToast('模型已创建', 'success')
+    }
+    editOpen.value = false
+    await load()
+  } catch (e) {
+    setToast(errMsg(e), 'error')
+  } finally {
+    saving.value = false
+  }
+}
+
+async function removeModel(m: Model) {
+  if (!confirm(`删除模型 ${m.name}？`)) return
+  try {
+    await request.delete(`/admin/models/${m.id}`)
+    setToast('模型已删除', 'success')
+    await load()
+  } catch (e) {
+    setToast(errMsg(e), 'error')
+  }
+}
+
+onMounted(load)
+</script>
+
+<template>
+  <div class="mx-auto max-w-6xl">
+    <div class="mb-6 flex flex-wrap items-center justify-between gap-3">
+      <div>
+        <h1 class="text-lg font-semibold">模型与定价</h1>
+        <p class="text-sm text-base-content/60">接口导入不全时可直接输入模型名添加，如 glm-4.7-flash</p>
+      </div>
+      <div class="flex w-full flex-wrap gap-2 sm:w-auto sm:flex-nowrap">
+        <input
+          v-model="quickName"
+          placeholder="模型名，如 glm-4.7-flash"
+          class="h-10 min-w-0 flex-1 rounded-md border border-base-300/60 bg-base-100 px-3 font-mono text-xs outline-none focus:border-primary sm:w-52 sm:flex-none"
+          @keyup.enter="quickAdd"
+        />
+        <Button class="shrink-0" @click="quickAdd">添加模型</Button>
+        <Button
+          size="md"
+          variant="danger"
+          class="shrink-0 px-1!"
+          :loading="clearing"
+          :disabled="!unused.length"
+          @click="clearUnused"
+        >
+          清除悬空{{ unused.length ? `（${unused.length}）` : '' }}
+        </Button>
+      </div>
+    </div>
+
+    <!-- 提示：定价目录 = 渠道选中的模型 + 手动添加的模型 -->
+    <div v-if="summary.missing.length" class="card border border-error/50 bg-error/5 p-4">
+      <p class="text-sm font-medium text-error">以下渠道选中的模型不在定价目录</p>
+      <p v-for="(x, i) in summary.missing" :key="i" class="mt-1 font-mono text-xs text-base-content/60">
+        {{ x.channel }} → {{ x.upstream_model || '模型 #' + x.model_id }}（请到渠道抽屉重新选中，或手动添加）
+      </p>
+    </div>
+    <p v-else-if="summary.unpriced > 0" class="text-xs text-base-content/60">
+      有 <span class="font-mono text-warning">{{ summary.unpriced }}</span> 个渠道允许的模型未定价，网关将按示例价计费
+    </p>
+    <p v-else class="text-xs text-base-content/60">定价目录中渠道允许的模型均已定价</p>
+
+    <div class="space-y-3">
+      <div v-for="m in models" :key="m.id" :class="m.channels.length ? 'card border border-base-300/60 bg-base-100' : 'card border border-warning/60 bg-warning/5'">
+        <div class="px-4 py-3">
+          <div class="flex flex-wrap items-center justify-between gap-x-4 gap-y-2">
+            <div class="flex flex-wrap items-center gap-2">
+              <span class="font-mono text-sm text-base-content">{{ m.name }}</span>
+              <Badge v-if="m.channels.length" variant="neutral">渠道允许</Badge>
+              <Badge v-else variant="warn">悬空</Badge>
+              <Badge :variant="m.enabled ? 'ok' : 'neutral'">{{ m.enabled ? '启用' : '停用' }}</Badge>
+              <Badge v-if="m.denied" variant="err">已禁止</Badge>
+              <Badge v-if="m.needs_pricing" variant="warn">未定价</Badge>
+            </div>
+            <div class="flex gap-2">
+              <button class="text-xs text-base-content/60 hover:text-base-content" @click="openEdit(m)">编辑</button>
+              <button class="text-xs text-base-content/60 hover:text-error" @click="removeModel(m)">删除</button>
+            </div>
+          </div>
+          <div class="mt-2 flex flex-wrap items-center gap-3">
+            <span class="font-mono text-xs text-base-content/60">入 {{ m.input_price }}</span>
+            <span class="font-mono text-xs text-base-content/60">出 {{ m.output_price }}</span>
+            <span class="font-mono text-xs text-base-content/60">缓存读 {{ m.cache_read_price }}</span>
+          </div>
+        </div>
+
+        <div v-if="m.channels.length" class="border-t border-base-300/60 px-4 py-2">
+          <p class="mb-1.5 text-[11px] font-medium text-base-content/50">允许渠道（渠道抽屉中管理）</p>
+          <div class="flex flex-wrap gap-2">
+            <span
+              v-for="b in m.channels"
+              :key="b.id"
+              class="inline-flex items-center rounded-md border border-base-300/60 bg-base-100 px-2 py-0.5 font-mono text-[11px] text-base-content/60"
+            >
+              {{ b.channel_name }} → {{ b.upstream_model }}
+            </span>
+          </div>
+        </div>
+        <p v-else class="border-t border-base-300/60 px-4 py-2 text-xs text-warning">
+          悬空模型：无任何渠道提供，客户端无法调用
+        </p>
+      </div>
+
+      <p v-if="models.length === 0" class="card border border-base-300/60 bg-base-100 px-4 py-10 text-center text-sm text-base-content/60">
+        还没有模型，点击「添加模型」或到渠道页「导入模型」
+      </p>
+    </div>
+
+    <!-- 模型编辑 -->
+    <Modal :open="editOpen" :title="editing ? '编辑模型' : '添加模型'" @close="editOpen = false">
+      <div class="space-y-4">
+        <Input v-model="form.name" label="模型名" placeholder="claude-sonnet-5" :disabled="!!editing" />
+        <div class="grid grid-cols-1 gap-4 sm:grid-cols-2">
+          <Input v-model="form.input_price" label="输入价格 /1M" type="number" />
+          <Input v-model="form.output_price" label="输出价格 /1M" type="number" />
+          <Input v-model="form.cache_read_price" label="缓存读价格 /1M" type="number" />
+        </div>
+      </div>
+      <template #footer>
+        <Button variant="ghost" @click="editOpen = false">取消</Button>
+        <Button :loading="saving" @click="save">{{ editing ? '保存' : '创建' }}</Button>
+      </template>
+    </Modal>
+  </div>
+</template>
diff --git a/frontend/src/views/dashboard/SystemConfig.vue b/frontend/src/views/dashboard/SystemConfig.vue
new file mode 100644
index 0000000..259bcea
--- /dev/null
+++ b/frontend/src/views/dashboard/SystemConfig.vue
@@ -0,0 +1,121 @@
+<script setup lang="ts">
+import { ref, onMounted } from 'vue'
+import request from '@/api/client'
+import { useToast } from '@/composables/toast'
+
+function errMsg(e: unknown) {
+  return (e as any)?.response?.data?.error || (e as any)?.message || '请求失败'
+}
+
+const { setToast } = useToast()
+const loading = ref(false)
+const saving = ref(false)
+
+const registrationEnabled = ref(true)
+const passwordLoginEnabled = ref(true)
+
+async function load() {
+  loading.value = true
+  try {
+    const [regRes, pwdRes] = await Promise.all([
+      request.get('/admin/config/registration'),
+      request.get('/admin/config/password-login'),
+    ])
+    registrationEnabled.value = regRes.data.data.enabled
+    passwordLoginEnabled.value = pwdRes.data.data.enabled
+  } catch (e) {
+    setToast(errMsg(e), 'error')
+  } finally {
+    loading.value = false
+  }
+}
+
+async function saveRegistration(enabled: boolean) {
+  saving.value = true
+  try {
+    await request.put('/admin/config/registration', { enabled })
+    registrationEnabled.value = enabled
+    setToast('注册设置已更新', 'success')
+  } catch (e) {
+    setToast(errMsg(e), 'error')
+  } finally {
+    saving.value = false
+  }
+}
+
+async function savePasswordLogin(enabled: boolean) {
+  saving.value = true
+  try {
+    await request.put('/admin/config/password-login', { enabled })
+    passwordLoginEnabled.value = enabled
+    setToast('密码登录设置已更新', 'success')
+  } catch (e) {
+    setToast(errMsg(e), 'error')
+  } finally {
+    saving.value = false
+  }
+}
+
+onMounted(load)
+</script>
+
+<template>
+  <div class="mx-auto max-w-2xl space-y-6">
+    <div>
+      <h1 class="text-lg font-semibold">系统配置</h1>
+      <p class="text-sm text-base-content/60">管理平台全局设置</p>
+    </div>
+
+    <div v-if="loading" class="py-10 text-center text-sm text-base-content/50">加载中…</div>
+
+    <template v-else>
+      <!-- 开放注册 -->
+      <div class="card border border-base-300/60 bg-base-100 p-4 shadow-sm">
+        <div class="flex items-center justify-between">
+          <div>
+            <h3 class="text-sm font-medium">开放注册</h3>
+            <p class="mt-1 text-xs text-base-content/50">允许新用户通过注册页面创建账号</p>
+          </div>
+          <button
+            type="button"
+            role="switch"
+            :aria-checked="registrationEnabled"
+            class="relative inline-flex h-6 w-11 shrink-0 cursor-pointer items-center rounded-full transition-colors focus-visible:outline-2 focus-visible:outline-offset-2 focus-visible:outline-primary"
+            :class="registrationEnabled ? 'bg-primary' : 'bg-base-200'"
+            :disabled="saving"
+            @click="saveRegistration(!registrationEnabled)"
+          >
+            <span
+              class="pointer-events-none inline-block h-4 w-4 rounded-full bg-white shadow-sm ring-0 transition-transform"
+              :class="registrationEnabled ? 'translate-x-6' : 'translate-x-1'"
+            />
+          </button>
+        </div>
+      </div>
+
+      <!-- 密码登录 -->
+      <div class="card border border-base-300/60 bg-base-100 p-4 shadow-sm">
+        <div class="flex items-center justify-between">
+          <div>
+            <h3 class="text-sm font-medium">密码登录</h3>
+            <p class="mt-1 text-xs text-base-content/50">允许用户通过用户名和密码登录</p>
+          </div>
+          <button
+            type="button"
+            role="switch"
+            :aria-checked="passwordLoginEnabled"
+            class="relative inline-flex h-6 w-11 shrink-0 cursor-pointer items-center rounded-full transition-colors focus-visible:outline-2 focus-visible:outline-offset-2 focus-visible:outline-primary"
+            :class="passwordLoginEnabled ? 'bg-primary' : 'bg-base-200'"
+            :disabled="saving"
+            @click="savePasswordLogin(!passwordLoginEnabled)"
+          >
+            <span
+              class="pointer-events-none inline-block h-4 w-4 rounded-full bg-white shadow-sm ring-0 transition-transform"
+              :class="passwordLoginEnabled ? 'translate-x-6' : 'translate-x-1'"
+            />
+          </button>
+        </div>
+      </div>
+    </template>
+  </div>
+</template>
diff --git a/frontend/vite.config.ts b/frontend/vite.config.ts
index 8333eb1..a30e534 100644
--- a/frontend/vite.config.ts
+++ b/frontend/vite.config.ts
@@ -9,7 +9,7 @@ import path from 'path'
 // 需要自签名 HTTPS 时设置 VITE_DEV_HTTPS=true
 const useHttps = process.env.VITE_DEV_HTTPS === 'true'
 // 后端地址：默认 make dev-backend 启动的 8080，可用 VITE_DEV_API_TARGET 覆盖
-const apiTarget = process.env.VITE_DEV_API_TARGET || 'http://localhost:8080'
+const apiTarget = process.env.VITE_DEV_API_TARGET || 'http://localhost:3000'
 
 // https://vite.dev/config/
 export default defineConfig({